1 /* -----------------------------------------------------------------------------
3 * (c) The GHC Team, 2000-2004
7 * ---------------------------------------------------------------------------*/
10 #include "PosixSource.h"
13 /* Linux needs _GNU_SOURCE to get RTLD_DEFAULT from <dlfcn.h> and
14 MREMAP_MAYMOVE from <sys/mman.h>.
23 #include "sm/Storage.h"
26 #include "LinkerInternals.h"
29 #include "StgPrimFloat.h" // for __int_encodeFloat etc.
32 #if !defined(mingw32_HOST_OS)
33 #include "posix/Signals.h"
36 #if defined(mingw32_HOST_OS)
37 // get protos for is*()
41 #ifdef HAVE_SYS_TYPES_H
42 #include <sys/types.h>
50 #ifdef HAVE_SYS_STAT_H
54 #if defined(HAVE_DLFCN_H)
58 #if defined(cygwin32_HOST_OS)
63 #ifdef HAVE_SYS_TIME_H
67 #include <sys/fcntl.h>
68 #include <sys/termios.h>
69 #include <sys/utime.h>
70 #include <sys/utsname.h>
74 #if defined(linux_HOST_OS) || defined(freebsd_HOST_OS) || defined(dragonfly_HOST_OS) || defined(netbsd_HOST_OS) || defined(openbsd_HOST_OS) || defined(darwin_HOST_OS)
85 #if defined(linux_HOST_OS) || defined(solaris2_HOST_OS) || defined(freebsd_HOST_OS) || defined(dragonfly_HOST_OS) || defined(netbsd_HOST_OS) || defined(openbsd_HOST_OS)
86 # define OBJFORMAT_ELF
87 # include <regex.h> // regex is already used by dlopen() so this is OK
88 // to use here without requiring an additional lib
89 #elif defined(cygwin32_HOST_OS) || defined (mingw32_HOST_OS)
90 # define OBJFORMAT_PEi386
93 #elif defined(darwin_HOST_OS)
94 # define OBJFORMAT_MACHO
96 # include <mach-o/loader.h>
97 # include <mach-o/nlist.h>
98 # include <mach-o/reloc.h>
99 #if !defined(HAVE_DLFCN_H)
100 # include <mach-o/dyld.h>
102 #if defined(powerpc_HOST_ARCH)
103 # include <mach-o/ppc/reloc.h>
105 #if defined(x86_64_HOST_ARCH)
106 # include <mach-o/x86_64/reloc.h>
110 #if defined(x86_64_HOST_ARCH) && defined(darwin_HOST_OS)
114 /* Hash table mapping symbol names to Symbol */
115 static /*Str*/HashTable *symhash;
117 /* Hash table mapping symbol names to StgStablePtr */
118 static /*Str*/HashTable *stablehash;
120 /* List of currently loaded objects */
121 ObjectCode *objects = NULL; /* initially empty */
123 #if defined(OBJFORMAT_ELF)
124 static int ocVerifyImage_ELF ( ObjectCode* oc );
125 static int ocGetNames_ELF ( ObjectCode* oc );
126 static int ocResolve_ELF ( ObjectCode* oc );
127 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
128 static int ocAllocateSymbolExtras_ELF ( ObjectCode* oc );
130 #elif defined(OBJFORMAT_PEi386)
131 static int ocVerifyImage_PEi386 ( ObjectCode* oc );
132 static int ocGetNames_PEi386 ( ObjectCode* oc );
133 static int ocResolve_PEi386 ( ObjectCode* oc );
134 static void *lookupSymbolInDLLs ( unsigned char *lbl );
135 static void zapTrailingAtSign ( unsigned char *sym );
136 #elif defined(OBJFORMAT_MACHO)
137 static int ocVerifyImage_MachO ( ObjectCode* oc );
138 static int ocGetNames_MachO ( ObjectCode* oc );
139 static int ocResolve_MachO ( ObjectCode* oc );
142 static int machoGetMisalignment( FILE * );
144 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
145 static int ocAllocateSymbolExtras_MachO ( ObjectCode* oc );
147 #ifdef powerpc_HOST_ARCH
148 static void machoInitSymbolsWithoutUnderscore( void );
152 /* on x86_64 we have a problem with relocating symbol references in
153 * code that was compiled without -fPIC. By default, the small memory
154 * model is used, which assumes that symbol references can fit in a
155 * 32-bit slot. The system dynamic linker makes this work for
156 * references to shared libraries by either (a) allocating a jump
157 * table slot for code references, or (b) moving the symbol at load
158 * time (and copying its contents, if necessary) for data references.
160 * We unfortunately can't tell whether symbol references are to code
161 * or data. So for now we assume they are code (the vast majority
162 * are), and allocate jump-table slots. Unfortunately this will
163 * SILENTLY generate crashing code for data references. This hack is
164 * enabled by X86_64_ELF_NONPIC_HACK.
166 * One workaround is to use shared Haskell libraries. This is
167 * coming. Another workaround is to keep the static libraries but
168 * compile them with -fPIC, because that will generate PIC references
169 * to data which can be relocated. The PIC code is still too green to
170 * do this systematically, though.
173 * See thread http://www.haskell.org/pipermail/cvs-ghc/2007-September/038458.html
175 * Naming Scheme for Symbol Macros
177 * SymI_*: symbol is internal to the RTS. It resides in an object
178 * file/library that is statically.
179 * SymE_*: symbol is external to the RTS library. It might be linked
182 * Sym*_HasProto : the symbol prototype is imported in an include file
183 * or defined explicitly
184 * Sym*_NeedsProto: the symbol is undefined and we add a dummy
185 * default proto extern void sym(void);
187 #define X86_64_ELF_NONPIC_HACK 1
189 /* Link objects into the lower 2Gb on x86_64. GHC assumes the
190 * small memory model on this architecture (see gcc docs,
193 * MAP_32BIT not available on OpenBSD/amd64
195 #if defined(x86_64_HOST_ARCH) && defined(MAP_32BIT)
196 #define TRY_MAP_32BIT MAP_32BIT
198 #define TRY_MAP_32BIT 0
202 * Due to the small memory model (see above), on x86_64 we have to map
203 * all our non-PIC object files into the low 2Gb of the address space
204 * (why 2Gb and not 4Gb? Because all addresses must be reachable
205 * using a 32-bit signed PC-relative offset). On Linux we can do this
206 * using the MAP_32BIT flag to mmap(), however on other OSs
207 * (e.g. *BSD, see #2063, and also on Linux inside Xen, see #2512), we
208 * can't do this. So on these systems, we have to pick a base address
209 * in the low 2Gb of the address space and try to allocate memory from
212 * We pick a default address based on the OS, but also make this
213 * configurable via an RTS flag (+RTS -xm)
215 #if !defined(ALWAYS_PIC) && defined(x86_64_HOST_ARCH)
217 #if defined(MAP_32BIT)
218 // Try to use MAP_32BIT
219 #define MMAP_32BIT_BASE_DEFAULT 0
222 #define MMAP_32BIT_BASE_DEFAULT 0x40000000
225 static void *mmap_32bit_base = (void *)MMAP_32BIT_BASE_DEFAULT;
228 /* MAP_ANONYMOUS is MAP_ANON on some systems, e.g. OpenBSD */
229 #if !defined(MAP_ANONYMOUS) && defined(MAP_ANON)
230 #define MAP_ANONYMOUS MAP_ANON
233 /* -----------------------------------------------------------------------------
234 * Built-in symbols from the RTS
237 typedef struct _RtsSymbolVal {
242 #define Maybe_Stable_Names SymI_HasProto(stg_mkWeakzh) \
243 SymI_HasProto(stg_mkWeakForeignEnvzh) \
244 SymI_HasProto(stg_makeStableNamezh) \
245 SymI_HasProto(stg_finalizzeWeakzh)
247 #if !defined (mingw32_HOST_OS)
248 #define RTS_POSIX_ONLY_SYMBOLS \
249 SymI_HasProto(__hscore_get_saved_termios) \
250 SymI_HasProto(__hscore_set_saved_termios) \
251 SymI_HasProto(shutdownHaskellAndSignal) \
252 SymI_HasProto(lockFile) \
253 SymI_HasProto(unlockFile) \
254 SymI_HasProto(signal_handlers) \
255 SymI_HasProto(stg_sig_install) \
256 SymI_NeedsProto(nocldstop)
259 #if defined (cygwin32_HOST_OS)
260 #define RTS_MINGW_ONLY_SYMBOLS /**/
261 /* Don't have the ability to read import libs / archives, so
262 * we have to stupidly list a lot of what libcygwin.a
265 #define RTS_CYGWIN_ONLY_SYMBOLS \
266 SymI_HasProto(regfree) \
267 SymI_HasProto(regexec) \
268 SymI_HasProto(regerror) \
269 SymI_HasProto(regcomp) \
270 SymI_HasProto(__errno) \
271 SymI_HasProto(access) \
272 SymI_HasProto(chmod) \
273 SymI_HasProto(chdir) \
274 SymI_HasProto(close) \
275 SymI_HasProto(creat) \
277 SymI_HasProto(dup2) \
278 SymI_HasProto(fstat) \
279 SymI_HasProto(fcntl) \
280 SymI_HasProto(getcwd) \
281 SymI_HasProto(getenv) \
282 SymI_HasProto(lseek) \
283 SymI_HasProto(open) \
284 SymI_HasProto(fpathconf) \
285 SymI_HasProto(pathconf) \
286 SymI_HasProto(stat) \
288 SymI_HasProto(tanh) \
289 SymI_HasProto(cosh) \
290 SymI_HasProto(sinh) \
291 SymI_HasProto(atan) \
292 SymI_HasProto(acos) \
293 SymI_HasProto(asin) \
299 SymI_HasProto(sqrt) \
300 SymI_HasProto(localtime_r) \
301 SymI_HasProto(gmtime_r) \
302 SymI_HasProto(mktime) \
303 SymI_NeedsProto(_imp___tzname) \
304 SymI_HasProto(gettimeofday) \
305 SymI_HasProto(timezone) \
306 SymI_HasProto(tcgetattr) \
307 SymI_HasProto(tcsetattr) \
308 SymI_HasProto(memcpy) \
309 SymI_HasProto(memmove) \
310 SymI_HasProto(realloc) \
311 SymI_HasProto(malloc) \
312 SymI_HasProto(free) \
313 SymI_HasProto(fork) \
314 SymI_HasProto(lstat) \
315 SymI_HasProto(isatty) \
316 SymI_HasProto(mkdir) \
317 SymI_HasProto(opendir) \
318 SymI_HasProto(readdir) \
319 SymI_HasProto(rewinddir) \
320 SymI_HasProto(closedir) \
321 SymI_HasProto(link) \
322 SymI_HasProto(mkfifo) \
323 SymI_HasProto(pipe) \
324 SymI_HasProto(read) \
325 SymI_HasProto(rename) \
326 SymI_HasProto(rmdir) \
327 SymI_HasProto(select) \
328 SymI_HasProto(system) \
329 SymI_HasProto(write) \
330 SymI_HasProto(strcmp) \
331 SymI_HasProto(strcpy) \
332 SymI_HasProto(strncpy) \
333 SymI_HasProto(strerror) \
334 SymI_HasProto(sigaddset) \
335 SymI_HasProto(sigemptyset) \
336 SymI_HasProto(sigprocmask) \
337 SymI_HasProto(umask) \
338 SymI_HasProto(uname) \
339 SymI_HasProto(unlink) \
340 SymI_HasProto(utime) \
341 SymI_HasProto(waitpid)
343 #elif !defined(mingw32_HOST_OS)
344 #define RTS_MINGW_ONLY_SYMBOLS /**/
345 #define RTS_CYGWIN_ONLY_SYMBOLS /**/
346 #else /* defined(mingw32_HOST_OS) */
347 #define RTS_POSIX_ONLY_SYMBOLS /**/
348 #define RTS_CYGWIN_ONLY_SYMBOLS /**/
350 /* Extra syms gen'ed by mingw-2's gcc-3.2: */
352 #define RTS_MINGW_EXTRA_SYMS \
353 SymI_NeedsProto(_imp____mb_cur_max) \
354 SymI_NeedsProto(_imp___pctype)
356 #define RTS_MINGW_EXTRA_SYMS
359 #if HAVE_GETTIMEOFDAY
360 #define RTS_MINGW_GETTIMEOFDAY_SYM SymI_NeedsProto(gettimeofday)
362 #define RTS_MINGW_GETTIMEOFDAY_SYM /**/
365 #if HAVE___MINGW_VFPRINTF
366 #define RTS___MINGW_VFPRINTF_SYM SymI_HasProto(__mingw_vfprintf)
368 #define RTS___MINGW_VFPRINTF_SYM /**/
371 /* These are statically linked from the mingw libraries into the ghc
372 executable, so we have to employ this hack. */
373 #define RTS_MINGW_ONLY_SYMBOLS \
374 SymI_HasProto(stg_asyncReadzh) \
375 SymI_HasProto(stg_asyncWritezh) \
376 SymI_HasProto(stg_asyncDoProczh) \
377 SymI_HasProto(memset) \
378 SymI_HasProto(inet_ntoa) \
379 SymI_HasProto(inet_addr) \
380 SymI_HasProto(htonl) \
381 SymI_HasProto(recvfrom) \
382 SymI_HasProto(listen) \
383 SymI_HasProto(bind) \
384 SymI_HasProto(shutdown) \
385 SymI_HasProto(connect) \
386 SymI_HasProto(htons) \
387 SymI_HasProto(ntohs) \
388 SymI_HasProto(getservbyname) \
389 SymI_HasProto(getservbyport) \
390 SymI_HasProto(getprotobynumber) \
391 SymI_HasProto(getprotobyname) \
392 SymI_HasProto(gethostbyname) \
393 SymI_HasProto(gethostbyaddr) \
394 SymI_HasProto(gethostname) \
395 SymI_HasProto(strcpy) \
396 SymI_HasProto(strncpy) \
397 SymI_HasProto(abort) \
398 SymI_NeedsProto(_alloca) \
399 SymI_HasProto(isxdigit) \
400 SymI_HasProto(isupper) \
401 SymI_HasProto(ispunct) \
402 SymI_HasProto(islower) \
403 SymI_HasProto(isspace) \
404 SymI_HasProto(isprint) \
405 SymI_HasProto(isdigit) \
406 SymI_HasProto(iscntrl) \
407 SymI_HasProto(isalpha) \
408 SymI_HasProto(isalnum) \
409 SymI_HasProto(isascii) \
410 RTS___MINGW_VFPRINTF_SYM \
411 SymI_HasProto(strcmp) \
412 SymI_HasProto(memmove) \
413 SymI_HasProto(realloc) \
414 SymI_HasProto(malloc) \
416 SymI_HasProto(tanh) \
417 SymI_HasProto(cosh) \
418 SymI_HasProto(sinh) \
419 SymI_HasProto(atan) \
420 SymI_HasProto(acos) \
421 SymI_HasProto(asin) \
427 SymI_HasProto(sqrt) \
428 SymI_HasProto(powf) \
429 SymI_HasProto(tanhf) \
430 SymI_HasProto(coshf) \
431 SymI_HasProto(sinhf) \
432 SymI_HasProto(atanf) \
433 SymI_HasProto(acosf) \
434 SymI_HasProto(asinf) \
435 SymI_HasProto(tanf) \
436 SymI_HasProto(cosf) \
437 SymI_HasProto(sinf) \
438 SymI_HasProto(expf) \
439 SymI_HasProto(logf) \
440 SymI_HasProto(sqrtf) \
442 SymI_HasProto(erfc) \
443 SymI_HasProto(erff) \
444 SymI_HasProto(erfcf) \
445 SymI_HasProto(memcpy) \
446 SymI_HasProto(rts_InstallConsoleEvent) \
447 SymI_HasProto(rts_ConsoleHandlerDone) \
448 SymI_NeedsProto(mktime) \
449 SymI_NeedsProto(_imp___timezone) \
450 SymI_NeedsProto(_imp___tzname) \
451 SymI_NeedsProto(_imp__tzname) \
452 SymI_NeedsProto(_imp___iob) \
453 SymI_NeedsProto(_imp___osver) \
454 SymI_NeedsProto(localtime) \
455 SymI_NeedsProto(gmtime) \
456 SymI_NeedsProto(opendir) \
457 SymI_NeedsProto(readdir) \
458 SymI_NeedsProto(rewinddir) \
459 RTS_MINGW_EXTRA_SYMS \
460 RTS_MINGW_GETTIMEOFDAY_SYM \
461 SymI_NeedsProto(closedir)
464 #if defined(darwin_HOST_OS) && HAVE_PRINTF_LDBLSTUB
465 #define RTS_DARWIN_ONLY_SYMBOLS \
466 SymI_NeedsProto(asprintf$LDBLStub) \
467 SymI_NeedsProto(err$LDBLStub) \
468 SymI_NeedsProto(errc$LDBLStub) \
469 SymI_NeedsProto(errx$LDBLStub) \
470 SymI_NeedsProto(fprintf$LDBLStub) \
471 SymI_NeedsProto(fscanf$LDBLStub) \
472 SymI_NeedsProto(fwprintf$LDBLStub) \
473 SymI_NeedsProto(fwscanf$LDBLStub) \
474 SymI_NeedsProto(printf$LDBLStub) \
475 SymI_NeedsProto(scanf$LDBLStub) \
476 SymI_NeedsProto(snprintf$LDBLStub) \
477 SymI_NeedsProto(sprintf$LDBLStub) \
478 SymI_NeedsProto(sscanf$LDBLStub) \
479 SymI_NeedsProto(strtold$LDBLStub) \
480 SymI_NeedsProto(swprintf$LDBLStub) \
481 SymI_NeedsProto(swscanf$LDBLStub) \
482 SymI_NeedsProto(syslog$LDBLStub) \
483 SymI_NeedsProto(vasprintf$LDBLStub) \
484 SymI_NeedsProto(verr$LDBLStub) \
485 SymI_NeedsProto(verrc$LDBLStub) \
486 SymI_NeedsProto(verrx$LDBLStub) \
487 SymI_NeedsProto(vfprintf$LDBLStub) \
488 SymI_NeedsProto(vfscanf$LDBLStub) \
489 SymI_NeedsProto(vfwprintf$LDBLStub) \
490 SymI_NeedsProto(vfwscanf$LDBLStub) \
491 SymI_NeedsProto(vprintf$LDBLStub) \
492 SymI_NeedsProto(vscanf$LDBLStub) \
493 SymI_NeedsProto(vsnprintf$LDBLStub) \
494 SymI_NeedsProto(vsprintf$LDBLStub) \
495 SymI_NeedsProto(vsscanf$LDBLStub) \
496 SymI_NeedsProto(vswprintf$LDBLStub) \
497 SymI_NeedsProto(vswscanf$LDBLStub) \
498 SymI_NeedsProto(vsyslog$LDBLStub) \
499 SymI_NeedsProto(vwarn$LDBLStub) \
500 SymI_NeedsProto(vwarnc$LDBLStub) \
501 SymI_NeedsProto(vwarnx$LDBLStub) \
502 SymI_NeedsProto(vwprintf$LDBLStub) \
503 SymI_NeedsProto(vwscanf$LDBLStub) \
504 SymI_NeedsProto(warn$LDBLStub) \
505 SymI_NeedsProto(warnc$LDBLStub) \
506 SymI_NeedsProto(warnx$LDBLStub) \
507 SymI_NeedsProto(wcstold$LDBLStub) \
508 SymI_NeedsProto(wprintf$LDBLStub) \
509 SymI_NeedsProto(wscanf$LDBLStub)
511 #define RTS_DARWIN_ONLY_SYMBOLS
515 # define MAIN_CAP_SYM SymI_HasProto(MainCapability)
517 # define MAIN_CAP_SYM
520 #if !defined(mingw32_HOST_OS)
521 #define RTS_USER_SIGNALS_SYMBOLS \
522 SymI_HasProto(setIOManagerControlFd) \
523 SymI_HasProto(setIOManagerWakeupFd) \
524 SymI_HasProto(ioManagerWakeup) \
525 SymI_HasProto(blockUserSignals) \
526 SymI_HasProto(unblockUserSignals)
528 #define RTS_USER_SIGNALS_SYMBOLS \
529 SymI_HasProto(ioManagerWakeup) \
530 SymI_HasProto(sendIOManagerEvent) \
531 SymI_HasProto(readIOManagerEvent) \
532 SymI_HasProto(getIOManagerEvent) \
533 SymI_HasProto(console_handler)
536 #define RTS_LIBFFI_SYMBOLS \
537 SymE_NeedsProto(ffi_prep_cif) \
538 SymE_NeedsProto(ffi_call) \
539 SymE_NeedsProto(ffi_type_void) \
540 SymE_NeedsProto(ffi_type_float) \
541 SymE_NeedsProto(ffi_type_double) \
542 SymE_NeedsProto(ffi_type_sint64) \
543 SymE_NeedsProto(ffi_type_uint64) \
544 SymE_NeedsProto(ffi_type_sint32) \
545 SymE_NeedsProto(ffi_type_uint32) \
546 SymE_NeedsProto(ffi_type_sint16) \
547 SymE_NeedsProto(ffi_type_uint16) \
548 SymE_NeedsProto(ffi_type_sint8) \
549 SymE_NeedsProto(ffi_type_uint8) \
550 SymE_NeedsProto(ffi_type_pointer)
552 #ifdef TABLES_NEXT_TO_CODE
553 #define RTS_RET_SYMBOLS /* nothing */
555 #define RTS_RET_SYMBOLS \
556 SymI_HasProto(stg_enter_ret) \
557 SymI_HasProto(stg_gc_fun_ret) \
558 SymI_HasProto(stg_ap_v_ret) \
559 SymI_HasProto(stg_ap_f_ret) \
560 SymI_HasProto(stg_ap_d_ret) \
561 SymI_HasProto(stg_ap_l_ret) \
562 SymI_HasProto(stg_ap_n_ret) \
563 SymI_HasProto(stg_ap_p_ret) \
564 SymI_HasProto(stg_ap_pv_ret) \
565 SymI_HasProto(stg_ap_pp_ret) \
566 SymI_HasProto(stg_ap_ppv_ret) \
567 SymI_HasProto(stg_ap_ppp_ret) \
568 SymI_HasProto(stg_ap_pppv_ret) \
569 SymI_HasProto(stg_ap_pppp_ret) \
570 SymI_HasProto(stg_ap_ppppp_ret) \
571 SymI_HasProto(stg_ap_pppppp_ret)
574 /* Modules compiled with -ticky may mention ticky counters */
575 /* This list should marry up with the one in $(TOP)/includes/stg/Ticky.h */
576 #define RTS_TICKY_SYMBOLS \
577 SymI_NeedsProto(ticky_entry_ctrs) \
578 SymI_NeedsProto(top_ct) \
580 SymI_HasProto(ENT_VIA_NODE_ctr) \
581 SymI_HasProto(ENT_STATIC_THK_ctr) \
582 SymI_HasProto(ENT_DYN_THK_ctr) \
583 SymI_HasProto(ENT_STATIC_FUN_DIRECT_ctr) \
584 SymI_HasProto(ENT_DYN_FUN_DIRECT_ctr) \
585 SymI_HasProto(ENT_STATIC_CON_ctr) \
586 SymI_HasProto(ENT_DYN_CON_ctr) \
587 SymI_HasProto(ENT_STATIC_IND_ctr) \
588 SymI_HasProto(ENT_DYN_IND_ctr) \
589 SymI_HasProto(ENT_PERM_IND_ctr) \
590 SymI_HasProto(ENT_PAP_ctr) \
591 SymI_HasProto(ENT_AP_ctr) \
592 SymI_HasProto(ENT_AP_STACK_ctr) \
593 SymI_HasProto(ENT_BH_ctr) \
594 SymI_HasProto(UNKNOWN_CALL_ctr) \
595 SymI_HasProto(SLOW_CALL_v_ctr) \
596 SymI_HasProto(SLOW_CALL_f_ctr) \
597 SymI_HasProto(SLOW_CALL_d_ctr) \
598 SymI_HasProto(SLOW_CALL_l_ctr) \
599 SymI_HasProto(SLOW_CALL_n_ctr) \
600 SymI_HasProto(SLOW_CALL_p_ctr) \
601 SymI_HasProto(SLOW_CALL_pv_ctr) \
602 SymI_HasProto(SLOW_CALL_pp_ctr) \
603 SymI_HasProto(SLOW_CALL_ppv_ctr) \
604 SymI_HasProto(SLOW_CALL_ppp_ctr) \
605 SymI_HasProto(SLOW_CALL_pppv_ctr) \
606 SymI_HasProto(SLOW_CALL_pppp_ctr) \
607 SymI_HasProto(SLOW_CALL_ppppp_ctr) \
608 SymI_HasProto(SLOW_CALL_pppppp_ctr) \
609 SymI_HasProto(SLOW_CALL_OTHER_ctr) \
610 SymI_HasProto(ticky_slow_call_unevald) \
611 SymI_HasProto(SLOW_CALL_ctr) \
612 SymI_HasProto(MULTI_CHUNK_SLOW_CALL_ctr) \
613 SymI_HasProto(MULTI_CHUNK_SLOW_CALL_CHUNKS_ctr) \
614 SymI_HasProto(KNOWN_CALL_ctr) \
615 SymI_HasProto(KNOWN_CALL_TOO_FEW_ARGS_ctr) \
616 SymI_HasProto(KNOWN_CALL_EXTRA_ARGS_ctr) \
617 SymI_HasProto(SLOW_CALL_FUN_TOO_FEW_ctr) \
618 SymI_HasProto(SLOW_CALL_FUN_CORRECT_ctr) \
619 SymI_HasProto(SLOW_CALL_FUN_TOO_MANY_ctr) \
620 SymI_HasProto(SLOW_CALL_PAP_TOO_FEW_ctr) \
621 SymI_HasProto(SLOW_CALL_PAP_CORRECT_ctr) \
622 SymI_HasProto(SLOW_CALL_PAP_TOO_MANY_ctr) \
623 SymI_HasProto(SLOW_CALL_UNEVALD_ctr) \
624 SymI_HasProto(UPDF_OMITTED_ctr) \
625 SymI_HasProto(UPDF_PUSHED_ctr) \
626 SymI_HasProto(CATCHF_PUSHED_ctr) \
627 SymI_HasProto(UPDF_RCC_PUSHED_ctr) \
628 SymI_HasProto(UPDF_RCC_OMITTED_ctr) \
629 SymI_HasProto(UPD_SQUEEZED_ctr) \
630 SymI_HasProto(UPD_CON_IN_NEW_ctr) \
631 SymI_HasProto(UPD_CON_IN_PLACE_ctr) \
632 SymI_HasProto(UPD_PAP_IN_NEW_ctr) \
633 SymI_HasProto(UPD_PAP_IN_PLACE_ctr) \
634 SymI_HasProto(ALLOC_HEAP_ctr) \
635 SymI_HasProto(ALLOC_HEAP_tot) \
636 SymI_HasProto(ALLOC_FUN_ctr) \
637 SymI_HasProto(ALLOC_FUN_adm) \
638 SymI_HasProto(ALLOC_FUN_gds) \
639 SymI_HasProto(ALLOC_FUN_slp) \
640 SymI_HasProto(UPD_NEW_IND_ctr) \
641 SymI_HasProto(UPD_NEW_PERM_IND_ctr) \
642 SymI_HasProto(UPD_OLD_IND_ctr) \
643 SymI_HasProto(UPD_OLD_PERM_IND_ctr) \
644 SymI_HasProto(UPD_BH_UPDATABLE_ctr) \
645 SymI_HasProto(UPD_BH_SINGLE_ENTRY_ctr) \
646 SymI_HasProto(UPD_CAF_BH_UPDATABLE_ctr) \
647 SymI_HasProto(UPD_CAF_BH_SINGLE_ENTRY_ctr) \
648 SymI_HasProto(GC_SEL_ABANDONED_ctr) \
649 SymI_HasProto(GC_SEL_MINOR_ctr) \
650 SymI_HasProto(GC_SEL_MAJOR_ctr) \
651 SymI_HasProto(GC_FAILED_PROMOTION_ctr) \
652 SymI_HasProto(ALLOC_UP_THK_ctr) \
653 SymI_HasProto(ALLOC_SE_THK_ctr) \
654 SymI_HasProto(ALLOC_THK_adm) \
655 SymI_HasProto(ALLOC_THK_gds) \
656 SymI_HasProto(ALLOC_THK_slp) \
657 SymI_HasProto(ALLOC_CON_ctr) \
658 SymI_HasProto(ALLOC_CON_adm) \
659 SymI_HasProto(ALLOC_CON_gds) \
660 SymI_HasProto(ALLOC_CON_slp) \
661 SymI_HasProto(ALLOC_TUP_ctr) \
662 SymI_HasProto(ALLOC_TUP_adm) \
663 SymI_HasProto(ALLOC_TUP_gds) \
664 SymI_HasProto(ALLOC_TUP_slp) \
665 SymI_HasProto(ALLOC_BH_ctr) \
666 SymI_HasProto(ALLOC_BH_adm) \
667 SymI_HasProto(ALLOC_BH_gds) \
668 SymI_HasProto(ALLOC_BH_slp) \
669 SymI_HasProto(ALLOC_PRIM_ctr) \
670 SymI_HasProto(ALLOC_PRIM_adm) \
671 SymI_HasProto(ALLOC_PRIM_gds) \
672 SymI_HasProto(ALLOC_PRIM_slp) \
673 SymI_HasProto(ALLOC_PAP_ctr) \
674 SymI_HasProto(ALLOC_PAP_adm) \
675 SymI_HasProto(ALLOC_PAP_gds) \
676 SymI_HasProto(ALLOC_PAP_slp) \
677 SymI_HasProto(ALLOC_TSO_ctr) \
678 SymI_HasProto(ALLOC_TSO_adm) \
679 SymI_HasProto(ALLOC_TSO_gds) \
680 SymI_HasProto(ALLOC_TSO_slp) \
681 SymI_HasProto(RET_NEW_ctr) \
682 SymI_HasProto(RET_OLD_ctr) \
683 SymI_HasProto(RET_UNBOXED_TUP_ctr) \
684 SymI_HasProto(RET_SEMI_loads_avoided)
687 // On most platforms, the garbage collector rewrites references
688 // to small integer and char objects to a set of common, shared ones.
690 // We don't do this when compiling to Windows DLLs at the moment because
691 // it doesn't support cross package data references well.
693 #if defined(__PIC__) && defined(mingw32_HOST_OS)
694 #define RTS_INTCHAR_SYMBOLS
696 #define RTS_INTCHAR_SYMBOLS \
697 SymI_HasProto(stg_CHARLIKE_closure) \
698 SymI_HasProto(stg_INTLIKE_closure)
702 #define RTS_SYMBOLS \
705 SymI_HasProto(StgReturn) \
706 SymI_HasProto(stg_enter_info) \
707 SymI_HasProto(stg_gc_void_info) \
708 SymI_HasProto(__stg_gc_enter_1) \
709 SymI_HasProto(stg_gc_noregs) \
710 SymI_HasProto(stg_gc_unpt_r1_info) \
711 SymI_HasProto(stg_gc_unpt_r1) \
712 SymI_HasProto(stg_gc_unbx_r1_info) \
713 SymI_HasProto(stg_gc_unbx_r1) \
714 SymI_HasProto(stg_gc_f1_info) \
715 SymI_HasProto(stg_gc_f1) \
716 SymI_HasProto(stg_gc_d1_info) \
717 SymI_HasProto(stg_gc_d1) \
718 SymI_HasProto(stg_gc_l1_info) \
719 SymI_HasProto(stg_gc_l1) \
720 SymI_HasProto(__stg_gc_fun) \
721 SymI_HasProto(stg_gc_fun_info) \
722 SymI_HasProto(stg_gc_gen) \
723 SymI_HasProto(stg_gc_gen_info) \
724 SymI_HasProto(stg_gc_gen_hp) \
725 SymI_HasProto(stg_gc_ut) \
726 SymI_HasProto(stg_gen_yield) \
727 SymI_HasProto(stg_yield_noregs) \
728 SymI_HasProto(stg_yield_to_interpreter) \
729 SymI_HasProto(stg_gen_block) \
730 SymI_HasProto(stg_block_noregs) \
731 SymI_HasProto(stg_block_1) \
732 SymI_HasProto(stg_block_takemvar) \
733 SymI_HasProto(stg_block_putmvar) \
735 SymI_HasProto(MallocFailHook) \
736 SymI_HasProto(OnExitHook) \
737 SymI_HasProto(OutOfHeapHook) \
738 SymI_HasProto(StackOverflowHook) \
739 SymI_HasProto(addDLL) \
740 SymI_HasProto(__int_encodeDouble) \
741 SymI_HasProto(__word_encodeDouble) \
742 SymI_HasProto(__2Int_encodeDouble) \
743 SymI_HasProto(__int_encodeFloat) \
744 SymI_HasProto(__word_encodeFloat) \
745 SymI_HasProto(stg_atomicallyzh) \
746 SymI_HasProto(barf) \
747 SymI_HasProto(debugBelch) \
748 SymI_HasProto(errorBelch) \
749 SymI_HasProto(sysErrorBelch) \
750 SymI_HasProto(stg_getMaskingStatezh) \
751 SymI_HasProto(stg_maskAsyncExceptionszh) \
752 SymI_HasProto(stg_maskUninterruptiblezh) \
753 SymI_HasProto(stg_catchzh) \
754 SymI_HasProto(stg_catchRetryzh) \
755 SymI_HasProto(stg_catchSTMzh) \
756 SymI_HasProto(stg_checkzh) \
757 SymI_HasProto(closure_flags) \
758 SymI_HasProto(cmp_thread) \
759 SymI_HasProto(createAdjustor) \
760 SymI_HasProto(stg_decodeDoublezu2Intzh) \
761 SymI_HasProto(stg_decodeFloatzuIntzh) \
762 SymI_HasProto(defaultsHook) \
763 SymI_HasProto(stg_delayzh) \
764 SymI_HasProto(stg_deRefWeakzh) \
765 SymI_HasProto(stg_deRefStablePtrzh) \
766 SymI_HasProto(dirty_MUT_VAR) \
767 SymI_HasProto(stg_forkzh) \
768 SymI_HasProto(stg_forkOnzh) \
769 SymI_HasProto(forkProcess) \
770 SymI_HasProto(forkOS_createThread) \
771 SymI_HasProto(freeHaskellFunctionPtr) \
772 SymI_HasProto(getOrSetTypeableStore) \
773 SymI_HasProto(getOrSetGHCConcSignalSignalHandlerStore) \
774 SymI_HasProto(getOrSetGHCConcWindowsPendingDelaysStore) \
775 SymI_HasProto(getOrSetGHCConcWindowsIOManagerThreadStore) \
776 SymI_HasProto(getOrSetGHCConcWindowsProddingStore) \
777 SymI_HasProto(getOrSetSystemEventThreadEventManagerStore) \
778 SymI_HasProto(getOrSetSystemEventThreadIOManagerThreadStore) \
779 SymI_HasProto(genSymZh) \
780 SymI_HasProto(genericRaise) \
781 SymI_HasProto(getProgArgv) \
782 SymI_HasProto(getFullProgArgv) \
783 SymI_HasProto(getStablePtr) \
784 SymI_HasProto(hs_init) \
785 SymI_HasProto(hs_exit) \
786 SymI_HasProto(hs_set_argv) \
787 SymI_HasProto(hs_add_root) \
788 SymI_HasProto(hs_perform_gc) \
789 SymI_HasProto(hs_free_stable_ptr) \
790 SymI_HasProto(hs_free_fun_ptr) \
791 SymI_HasProto(hs_hpc_rootModule) \
792 SymI_HasProto(hs_hpc_module) \
793 SymI_HasProto(initLinker) \
794 SymI_HasProto(stg_unpackClosurezh) \
795 SymI_HasProto(stg_getApStackValzh) \
796 SymI_HasProto(stg_getSparkzh) \
797 SymI_HasProto(stg_numSparkszh) \
798 SymI_HasProto(stg_isCurrentThreadBoundzh) \
799 SymI_HasProto(stg_isEmptyMVarzh) \
800 SymI_HasProto(stg_killThreadzh) \
801 SymI_HasProto(loadObj) \
802 SymI_HasProto(insertStableSymbol) \
803 SymI_HasProto(insertSymbol) \
804 SymI_HasProto(lookupSymbol) \
805 SymI_HasProto(stg_makeStablePtrzh) \
806 SymI_HasProto(stg_mkApUpd0zh) \
807 SymI_HasProto(stg_myThreadIdzh) \
808 SymI_HasProto(stg_labelThreadzh) \
809 SymI_HasProto(stg_newArrayzh) \
810 SymI_HasProto(stg_newBCOzh) \
811 SymI_HasProto(stg_newByteArrayzh) \
812 SymI_HasProto_redirect(newCAF, newDynCAF) \
813 SymI_HasProto(stg_newMVarzh) \
814 SymI_HasProto(stg_newMutVarzh) \
815 SymI_HasProto(stg_newTVarzh) \
816 SymI_HasProto(stg_noDuplicatezh) \
817 SymI_HasProto(stg_atomicModifyMutVarzh) \
818 SymI_HasProto(stg_newPinnedByteArrayzh) \
819 SymI_HasProto(stg_newAlignedPinnedByteArrayzh) \
820 SymI_HasProto(newSpark) \
821 SymI_HasProto(performGC) \
822 SymI_HasProto(performMajorGC) \
823 SymI_HasProto(prog_argc) \
824 SymI_HasProto(prog_argv) \
825 SymI_HasProto(stg_putMVarzh) \
826 SymI_HasProto(stg_raisezh) \
827 SymI_HasProto(stg_raiseIOzh) \
828 SymI_HasProto(stg_readTVarzh) \
829 SymI_HasProto(stg_readTVarIOzh) \
830 SymI_HasProto(resumeThread) \
831 SymI_HasProto(resolveObjs) \
832 SymI_HasProto(stg_retryzh) \
833 SymI_HasProto(rts_apply) \
834 SymI_HasProto(rts_checkSchedStatus) \
835 SymI_HasProto(rts_eval) \
836 SymI_HasProto(rts_evalIO) \
837 SymI_HasProto(rts_evalLazyIO) \
838 SymI_HasProto(rts_evalStableIO) \
839 SymI_HasProto(rts_eval_) \
840 SymI_HasProto(rts_getBool) \
841 SymI_HasProto(rts_getChar) \
842 SymI_HasProto(rts_getDouble) \
843 SymI_HasProto(rts_getFloat) \
844 SymI_HasProto(rts_getInt) \
845 SymI_HasProto(rts_getInt8) \
846 SymI_HasProto(rts_getInt16) \
847 SymI_HasProto(rts_getInt32) \
848 SymI_HasProto(rts_getInt64) \
849 SymI_HasProto(rts_getPtr) \
850 SymI_HasProto(rts_getFunPtr) \
851 SymI_HasProto(rts_getStablePtr) \
852 SymI_HasProto(rts_getThreadId) \
853 SymI_HasProto(rts_getWord) \
854 SymI_HasProto(rts_getWord8) \
855 SymI_HasProto(rts_getWord16) \
856 SymI_HasProto(rts_getWord32) \
857 SymI_HasProto(rts_getWord64) \
858 SymI_HasProto(rts_lock) \
859 SymI_HasProto(rts_mkBool) \
860 SymI_HasProto(rts_mkChar) \
861 SymI_HasProto(rts_mkDouble) \
862 SymI_HasProto(rts_mkFloat) \
863 SymI_HasProto(rts_mkInt) \
864 SymI_HasProto(rts_mkInt8) \
865 SymI_HasProto(rts_mkInt16) \
866 SymI_HasProto(rts_mkInt32) \
867 SymI_HasProto(rts_mkInt64) \
868 SymI_HasProto(rts_mkPtr) \
869 SymI_HasProto(rts_mkFunPtr) \
870 SymI_HasProto(rts_mkStablePtr) \
871 SymI_HasProto(rts_mkString) \
872 SymI_HasProto(rts_mkWord) \
873 SymI_HasProto(rts_mkWord8) \
874 SymI_HasProto(rts_mkWord16) \
875 SymI_HasProto(rts_mkWord32) \
876 SymI_HasProto(rts_mkWord64) \
877 SymI_HasProto(rts_unlock) \
878 SymI_HasProto(rts_unsafeGetMyCapability) \
879 SymI_HasProto(rtsSupportsBoundThreads) \
880 SymI_HasProto(setProgArgv) \
881 SymI_HasProto(startupHaskell) \
882 SymI_HasProto(shutdownHaskell) \
883 SymI_HasProto(shutdownHaskellAndExit) \
884 SymI_HasProto(stable_ptr_table) \
885 SymI_HasProto(stackOverflow) \
886 SymI_HasProto(stg_CAF_BLACKHOLE_info) \
887 SymI_HasProto(stg_BLACKHOLE_info) \
888 SymI_HasProto(__stg_EAGER_BLACKHOLE_info) \
889 SymI_HasProto(stg_BLOCKING_QUEUE_CLEAN_info) \
890 SymI_HasProto(stg_BLOCKING_QUEUE_DIRTY_info) \
891 SymI_HasProto(startTimer) \
892 SymI_HasProto(stg_MVAR_CLEAN_info) \
893 SymI_HasProto(stg_MVAR_DIRTY_info) \
894 SymI_HasProto(stg_IND_STATIC_info) \
895 SymI_HasProto(stg_ARR_WORDS_info) \
896 SymI_HasProto(stg_MUT_ARR_PTRS_DIRTY_info) \
897 SymI_HasProto(stg_MUT_ARR_PTRS_FROZEN_info) \
898 SymI_HasProto(stg_MUT_ARR_PTRS_FROZEN0_info) \
899 SymI_HasProto(stg_WEAK_info) \
900 SymI_HasProto(stg_ap_v_info) \
901 SymI_HasProto(stg_ap_f_info) \
902 SymI_HasProto(stg_ap_d_info) \
903 SymI_HasProto(stg_ap_l_info) \
904 SymI_HasProto(stg_ap_n_info) \
905 SymI_HasProto(stg_ap_p_info) \
906 SymI_HasProto(stg_ap_pv_info) \
907 SymI_HasProto(stg_ap_pp_info) \
908 SymI_HasProto(stg_ap_ppv_info) \
909 SymI_HasProto(stg_ap_ppp_info) \
910 SymI_HasProto(stg_ap_pppv_info) \
911 SymI_HasProto(stg_ap_pppp_info) \
912 SymI_HasProto(stg_ap_ppppp_info) \
913 SymI_HasProto(stg_ap_pppppp_info) \
914 SymI_HasProto(stg_ap_0_fast) \
915 SymI_HasProto(stg_ap_v_fast) \
916 SymI_HasProto(stg_ap_f_fast) \
917 SymI_HasProto(stg_ap_d_fast) \
918 SymI_HasProto(stg_ap_l_fast) \
919 SymI_HasProto(stg_ap_n_fast) \
920 SymI_HasProto(stg_ap_p_fast) \
921 SymI_HasProto(stg_ap_pv_fast) \
922 SymI_HasProto(stg_ap_pp_fast) \
923 SymI_HasProto(stg_ap_ppv_fast) \
924 SymI_HasProto(stg_ap_ppp_fast) \
925 SymI_HasProto(stg_ap_pppv_fast) \
926 SymI_HasProto(stg_ap_pppp_fast) \
927 SymI_HasProto(stg_ap_ppppp_fast) \
928 SymI_HasProto(stg_ap_pppppp_fast) \
929 SymI_HasProto(stg_ap_1_upd_info) \
930 SymI_HasProto(stg_ap_2_upd_info) \
931 SymI_HasProto(stg_ap_3_upd_info) \
932 SymI_HasProto(stg_ap_4_upd_info) \
933 SymI_HasProto(stg_ap_5_upd_info) \
934 SymI_HasProto(stg_ap_6_upd_info) \
935 SymI_HasProto(stg_ap_7_upd_info) \
936 SymI_HasProto(stg_exit) \
937 SymI_HasProto(stg_sel_0_upd_info) \
938 SymI_HasProto(stg_sel_10_upd_info) \
939 SymI_HasProto(stg_sel_11_upd_info) \
940 SymI_HasProto(stg_sel_12_upd_info) \
941 SymI_HasProto(stg_sel_13_upd_info) \
942 SymI_HasProto(stg_sel_14_upd_info) \
943 SymI_HasProto(stg_sel_15_upd_info) \
944 SymI_HasProto(stg_sel_1_upd_info) \
945 SymI_HasProto(stg_sel_2_upd_info) \
946 SymI_HasProto(stg_sel_3_upd_info) \
947 SymI_HasProto(stg_sel_4_upd_info) \
948 SymI_HasProto(stg_sel_5_upd_info) \
949 SymI_HasProto(stg_sel_6_upd_info) \
950 SymI_HasProto(stg_sel_7_upd_info) \
951 SymI_HasProto(stg_sel_8_upd_info) \
952 SymI_HasProto(stg_sel_9_upd_info) \
953 SymI_HasProto(stg_upd_frame_info) \
954 SymI_HasProto(stg_bh_upd_frame_info) \
955 SymI_HasProto(suspendThread) \
956 SymI_HasProto(stg_takeMVarzh) \
957 SymI_HasProto(stg_threadStatuszh) \
958 SymI_HasProto(stg_tryPutMVarzh) \
959 SymI_HasProto(stg_tryTakeMVarzh) \
960 SymI_HasProto(stg_unmaskAsyncExceptionszh) \
961 SymI_HasProto(unloadObj) \
962 SymI_HasProto(stg_unsafeThawArrayzh) \
963 SymI_HasProto(stg_waitReadzh) \
964 SymI_HasProto(stg_waitWritezh) \
965 SymI_HasProto(stg_writeTVarzh) \
966 SymI_HasProto(stg_yieldzh) \
967 SymI_NeedsProto(stg_interp_constr_entry) \
968 SymI_HasProto(alloc_blocks_lim) \
970 SymI_HasProto(allocate) \
971 SymI_HasProto(allocateExec) \
972 SymI_HasProto(freeExec) \
973 SymI_HasProto(getAllocations) \
974 SymI_HasProto(revertCAFs) \
975 SymI_HasProto(RtsFlags) \
976 SymI_NeedsProto(rts_breakpoint_io_action) \
977 SymI_NeedsProto(rts_stop_next_breakpoint) \
978 SymI_NeedsProto(rts_stop_on_exception) \
979 SymI_HasProto(stopTimer) \
980 SymI_HasProto(n_capabilities) \
981 SymI_HasProto(stg_traceCcszh) \
982 SymI_HasProto(stg_traceEventzh) \
983 RTS_USER_SIGNALS_SYMBOLS \
987 // 64-bit support functions in libgcc.a
988 #if defined(__GNUC__) && SIZEOF_VOID_P <= 4
989 #define RTS_LIBGCC_SYMBOLS \
990 SymI_NeedsProto(__divdi3) \
991 SymI_NeedsProto(__udivdi3) \
992 SymI_NeedsProto(__moddi3) \
993 SymI_NeedsProto(__umoddi3) \
994 SymI_NeedsProto(__muldi3) \
995 SymI_NeedsProto(__ashldi3) \
996 SymI_NeedsProto(__ashrdi3) \
997 SymI_NeedsProto(__lshrdi3)
999 #define RTS_LIBGCC_SYMBOLS
1002 #if defined(darwin_HOST_OS) && defined(powerpc_HOST_ARCH)
1003 // Symbols that don't have a leading underscore
1004 // on Mac OS X. They have to receive special treatment,
1005 // see machoInitSymbolsWithoutUnderscore()
1006 #define RTS_MACHO_NOUNDERLINE_SYMBOLS \
1007 SymI_NeedsProto(saveFP) \
1008 SymI_NeedsProto(restFP)
1011 /* entirely bogus claims about types of these symbols */
1012 #define SymI_NeedsProto(vvv) extern void vvv(void);
1013 #if defined(__PIC__) && defined(mingw32_HOST_OS)
1014 #define SymE_HasProto(vvv) SymE_HasProto(vvv);
1015 #define SymE_NeedsProto(vvv) extern void _imp__ ## vvv (void);
1017 #define SymE_NeedsProto(vvv) SymI_NeedsProto(vvv);
1018 #define SymE_HasProto(vvv) SymI_HasProto(vvv)
1020 #define SymI_HasProto(vvv) /**/
1021 #define SymI_HasProto_redirect(vvv,xxx) /**/
1024 RTS_POSIX_ONLY_SYMBOLS
1025 RTS_MINGW_ONLY_SYMBOLS
1026 RTS_CYGWIN_ONLY_SYMBOLS
1027 RTS_DARWIN_ONLY_SYMBOLS
1030 #undef SymI_NeedsProto
1031 #undef SymI_HasProto
1032 #undef SymI_HasProto_redirect
1033 #undef SymE_HasProto
1034 #undef SymE_NeedsProto
1036 #ifdef LEADING_UNDERSCORE
1037 #define MAYBE_LEADING_UNDERSCORE_STR(s) ("_" s)
1039 #define MAYBE_LEADING_UNDERSCORE_STR(s) (s)
1042 #define SymI_HasProto(vvv) { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
1044 #define SymE_HasProto(vvv) { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
1045 (void*)DLL_IMPORT_DATA_REF(vvv) },
1047 #define SymI_NeedsProto(vvv) SymI_HasProto(vvv)
1048 #define SymE_NeedsProto(vvv) SymE_HasProto(vvv)
1050 // SymI_HasProto_redirect allows us to redirect references to one symbol to
1051 // another symbol. See newCAF/newDynCAF for an example.
1052 #define SymI_HasProto_redirect(vvv,xxx) \
1053 { MAYBE_LEADING_UNDERSCORE_STR(#vvv), \
1056 static RtsSymbolVal rtsSyms[] = {
1059 RTS_POSIX_ONLY_SYMBOLS
1060 RTS_MINGW_ONLY_SYMBOLS
1061 RTS_CYGWIN_ONLY_SYMBOLS
1062 RTS_DARWIN_ONLY_SYMBOLS
1065 #if defined(darwin_HOST_OS) && defined(i386_HOST_ARCH)
1066 // dyld stub code contains references to this,
1067 // but it should never be called because we treat
1068 // lazy pointers as nonlazy.
1069 { "dyld_stub_binding_helper", (void*)0xDEADBEEF },
1071 { 0, 0 } /* sentinel */
1076 /* -----------------------------------------------------------------------------
1077 * Insert symbols into hash tables, checking for duplicates.
1080 static void ghciInsertStrHashTable ( char* obj_name,
1086 if (lookupHashTable(table, (StgWord)key) == NULL)
1088 insertStrHashTable(table, (StgWord)key, data);
1093 "GHCi runtime linker: fatal error: I found a duplicate definition for symbol\n"
1095 "whilst processing object file\n"
1097 "This could be caused by:\n"
1098 " * Loading two different object files which export the same symbol\n"
1099 " * Specifying the same object file twice on the GHCi command line\n"
1100 " * An incorrect `package.conf' entry, causing some object to be\n"
1102 "GHCi cannot safely continue in this situation. Exiting now. Sorry.\n"
1109 /* -----------------------------------------------------------------------------
1110 * initialize the object linker
1114 static int linker_init_done = 0 ;
1116 #if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1117 static void *dl_prog_handle;
1118 static regex_t re_invalid;
1119 static regex_t re_realso;
1121 static Mutex dl_mutex; // mutex to protect dlopen/dlerror critical section
1129 #if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1133 /* Make initLinker idempotent, so we can call it
1134 before evey relevant operation; that means we
1135 don't need to initialise the linker separately */
1136 if (linker_init_done == 1) { return; } else {
1137 linker_init_done = 1;
1140 #if defined(THREADED_RTS) && (defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO))
1141 initMutex(&dl_mutex);
1143 stablehash = allocStrHashTable();
1144 symhash = allocStrHashTable();
1146 /* populate the symbol table with stuff from the RTS */
1147 for (sym = rtsSyms; sym->lbl != NULL; sym++) {
1148 ghciInsertStrHashTable("(GHCi built-in symbols)",
1149 symhash, sym->lbl, sym->addr);
1151 # if defined(OBJFORMAT_MACHO) && defined(powerpc_HOST_ARCH)
1152 machoInitSymbolsWithoutUnderscore();
1155 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1156 # if defined(RTLD_DEFAULT)
1157 dl_prog_handle = RTLD_DEFAULT;
1159 dl_prog_handle = dlopen(NULL, RTLD_LAZY);
1160 # endif /* RTLD_DEFAULT */
1162 compileResult = regcomp(&re_invalid,
1163 "(([^ \t()])+\\.so([^ \t:()])*):([ \t])*invalid ELF header",
1165 ASSERT( compileResult == 0 );
1166 compileResult = regcomp(&re_realso,
1167 "GROUP *\\( *(([^ )])+)",
1169 ASSERT( compileResult == 0 );
1172 #if !defined(ALWAYS_PIC) && defined(x86_64_HOST_ARCH)
1173 if (RtsFlags.MiscFlags.linkerMemBase != 0) {
1174 // User-override for mmap_32bit_base
1175 mmap_32bit_base = (void*)RtsFlags.MiscFlags.linkerMemBase;
1179 #if defined(mingw32_HOST_OS)
1181 * These two libraries cause problems when added to the static link,
1182 * but are necessary for resolving symbols in GHCi, hence we load
1183 * them manually here.
1191 exitLinker( void ) {
1192 #if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1193 if (linker_init_done == 1) {
1194 regfree(&re_invalid);
1195 regfree(&re_realso);
1197 closeMutex(&dl_mutex);
1203 /* -----------------------------------------------------------------------------
1204 * Loading DLL or .so dynamic libraries
1205 * -----------------------------------------------------------------------------
1207 * Add a DLL from which symbols may be found. In the ELF case, just
1208 * do RTLD_GLOBAL-style add, so no further messing around needs to
1209 * happen in order that symbols in the loaded .so are findable --
1210 * lookupSymbol() will subsequently see them by dlsym on the program's
1211 * dl-handle. Returns NULL if success, otherwise ptr to an err msg.
1213 * In the PEi386 case, open the DLLs and put handles to them in a
1214 * linked list. When looking for a symbol, try all handles in the
1215 * list. This means that we need to load even DLLs that are guaranteed
1216 * to be in the ghc.exe image already, just so we can get a handle
1217 * to give to loadSymbol, so that we can find the symbols. For such
1218 * libraries, the LoadLibrary call should be a no-op except for returning
1223 #if defined(OBJFORMAT_PEi386)
1224 /* A record for storing handles into DLLs. */
1229 struct _OpenedDLL* next;
1234 /* A list thereof. */
1235 static OpenedDLL* opened_dlls = NULL;
1238 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1241 internal_dlopen(const char *dll_name)
1247 // omitted: RTLD_NOW
1248 // see http://www.haskell.org/pipermail/cvs-ghc/2007-September/038570.html
1250 debugBelch("internal_dlopen: dll_name = '%s'\n", dll_name));
1252 //-------------- Begin critical section ------------------
1253 // This critical section is necessary because dlerror() is not
1254 // required to be reentrant (see POSIX -- IEEE Std 1003.1-2008)
1255 // Also, the error message returned must be copied to preserve it
1258 ACQUIRE_LOCK(&dl_mutex);
1259 hdl = dlopen(dll_name, RTLD_LAZY | RTLD_GLOBAL);
1263 /* dlopen failed; return a ptr to the error msg. */
1265 if (errmsg == NULL) errmsg = "addDLL: unknown error";
1266 errmsg_copy = stgMallocBytes(strlen(errmsg)+1, "addDLL");
1267 strcpy(errmsg_copy, errmsg);
1268 errmsg = errmsg_copy;
1270 RELEASE_LOCK(&dl_mutex);
1271 //--------------- End critical section -------------------
1278 addDLL( char *dll_name )
1280 # if defined(OBJFORMAT_ELF) || defined(OBJFORMAT_MACHO)
1281 /* ------------------- ELF DLL loader ------------------- */
1284 regmatch_t match[NMATCH];
1287 size_t match_length;
1288 #define MAXLINE 1000
1294 IF_DEBUG(linker, debugBelch("addDLL: dll_name = '%s'\n", dll_name));
1295 errmsg = internal_dlopen(dll_name);
1297 if (errmsg == NULL) {
1301 // GHC Trac ticket #2615
1302 // On some systems (e.g., Gentoo Linux) dynamic files (e.g. libc.so)
1303 // contain linker scripts rather than ELF-format object code. This
1304 // code handles the situation by recognizing the real object code
1305 // file name given in the linker script.
1307 // If an "invalid ELF header" error occurs, it is assumed that the
1308 // .so file contains a linker script instead of ELF object code.
1309 // In this case, the code looks for the GROUP ( ... ) linker
1310 // directive. If one is found, the first file name inside the
1311 // parentheses is treated as the name of a dynamic library and the
1312 // code attempts to dlopen that file. If this is also unsuccessful,
1313 // an error message is returned.
1315 // see if the error message is due to an invalid ELF header
1316 IF_DEBUG(linker, debugBelch("errmsg = '%s'\n", errmsg));
1317 result = regexec(&re_invalid, errmsg, (size_t) NMATCH, match, 0);
1318 IF_DEBUG(linker, debugBelch("result = %i\n", result));
1320 // success -- try to read the named file as a linker script
1321 match_length = (size_t) stg_min((match[1].rm_eo - match[1].rm_so),
1323 strncpy(line, (errmsg+(match[1].rm_so)),match_length);
1324 line[match_length] = '\0'; // make sure string is null-terminated
1325 IF_DEBUG(linker, debugBelch ("file name = '%s'\n", line));
1326 if ((fp = fopen(line, "r")) == NULL) {
1327 return errmsg; // return original error if open fails
1329 // try to find a GROUP ( ... ) command
1330 while (fgets(line, MAXLINE, fp) != NULL) {
1331 IF_DEBUG(linker, debugBelch("input line = %s", line));
1332 if (regexec(&re_realso, line, (size_t) NMATCH, match, 0) == 0) {
1333 // success -- try to dlopen the first named file
1334 IF_DEBUG(linker, debugBelch("match%s\n",""));
1335 line[match[1].rm_eo] = '\0';
1336 errmsg = internal_dlopen(line+match[1].rm_so);
1339 // if control reaches here, no GROUP ( ... ) directive was found
1340 // and the original error message is returned to the caller
1346 # elif defined(OBJFORMAT_PEi386)
1347 /* ------------------- Win32 DLL loader ------------------- */
1355 /* debugBelch("\naddDLL; dll_name = `%s'\n", dll_name); */
1357 /* See if we've already got it, and ignore if so. */
1358 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
1359 if (0 == strcmp(o_dll->name, dll_name))
1363 /* The file name has no suffix (yet) so that we can try
1364 both foo.dll and foo.drv
1366 The documentation for LoadLibrary says:
1367 If no file name extension is specified in the lpFileName
1368 parameter, the default library extension .dll is
1369 appended. However, the file name string can include a trailing
1370 point character (.) to indicate that the module name has no
1373 buf = stgMallocBytes(strlen(dll_name) + 10, "addDLL");
1374 sprintf(buf, "%s.DLL", dll_name);
1375 instance = LoadLibrary(buf);
1376 if (instance == NULL) {
1377 if (GetLastError() != ERROR_MOD_NOT_FOUND) goto error;
1378 // KAA: allow loading of drivers (like winspool.drv)
1379 sprintf(buf, "%s.DRV", dll_name);
1380 instance = LoadLibrary(buf);
1381 if (instance == NULL) {
1382 if (GetLastError() != ERROR_MOD_NOT_FOUND) goto error;
1383 // #1883: allow loading of unix-style libfoo.dll DLLs
1384 sprintf(buf, "lib%s.DLL", dll_name);
1385 instance = LoadLibrary(buf);
1386 if (instance == NULL) {
1393 /* Add this DLL to the list of DLLs in which to search for symbols. */
1394 o_dll = stgMallocBytes( sizeof(OpenedDLL), "addDLL" );
1395 o_dll->name = stgMallocBytes(1+strlen(dll_name), "addDLL");
1396 strcpy(o_dll->name, dll_name);
1397 o_dll->instance = instance;
1398 o_dll->next = opened_dlls;
1399 opened_dlls = o_dll;
1405 sysErrorBelch(dll_name);
1407 /* LoadLibrary failed; return a ptr to the error msg. */
1408 return "addDLL: could not load DLL";
1411 barf("addDLL: not implemented on this platform");
1415 /* -----------------------------------------------------------------------------
1416 * insert a stable symbol in the hash table
1420 insertStableSymbol(char* obj_name, char* key, StgPtr p)
1422 ghciInsertStrHashTable(obj_name, stablehash, key, getStablePtr(p));
1426 /* -----------------------------------------------------------------------------
1427 * insert a symbol in the hash table
1430 insertSymbol(char* obj_name, char* key, void* data)
1432 ghciInsertStrHashTable(obj_name, symhash, key, data);
1435 /* -----------------------------------------------------------------------------
1436 * lookup a symbol in the hash table
1439 lookupSymbol( char *lbl )
1443 ASSERT(symhash != NULL);
1444 val = lookupStrHashTable(symhash, lbl);
1447 # if defined(OBJFORMAT_ELF)
1448 return dlsym(dl_prog_handle, lbl);
1449 # elif defined(OBJFORMAT_MACHO)
1451 /* On OS X 10.3 and later, we use dlsym instead of the old legacy
1454 HACK: On OS X, global symbols are prefixed with an underscore.
1455 However, dlsym wants us to omit the leading underscore from the
1456 symbol name. For now, we simply strip it off here (and ONLY
1459 ASSERT(lbl[0] == '_');
1460 return dlsym(dl_prog_handle, lbl+1);
1462 if(NSIsSymbolNameDefined(lbl)) {
1463 NSSymbol symbol = NSLookupAndBindSymbol(lbl);
1464 return NSAddressOfSymbol(symbol);
1468 # endif /* HAVE_DLFCN_H */
1469 # elif defined(OBJFORMAT_PEi386)
1472 sym = lookupSymbolInDLLs((unsigned char*)lbl);
1473 if (sym != NULL) { return sym; };
1475 // Also try looking up the symbol without the @N suffix. Some
1476 // DLLs have the suffixes on their symbols, some don't.
1477 zapTrailingAtSign ( (unsigned char*)lbl );
1478 sym = lookupSymbolInDLLs((unsigned char*)lbl);
1479 if (sym != NULL) { return sym; };
1491 /* -----------------------------------------------------------------------------
1492 * Debugging aid: look in GHCi's object symbol tables for symbols
1493 * within DELTA bytes of the specified address, and show their names.
1496 void ghci_enquire ( char* addr );
1498 void ghci_enquire ( char* addr )
1503 const int DELTA = 64;
1508 for (oc = objects; oc; oc = oc->next) {
1509 for (i = 0; i < oc->n_symbols; i++) {
1510 sym = oc->symbols[i];
1511 if (sym == NULL) continue;
1514 a = lookupStrHashTable(symhash, sym);
1517 // debugBelch("ghci_enquire: can't find %s\n", sym);
1519 else if (addr-DELTA <= a && a <= addr+DELTA) {
1520 debugBelch("%p + %3d == `%s'\n", addr, (int)(a - addr), sym);
1528 #define ROUND_UP(x,size) ((x + size - 1) & ~(size - 1))
1531 mmapForLinker (size_t bytes, nat flags, int fd)
1533 void *map_addr = NULL;
1536 static nat fixed = 0;
1538 pagesize = getpagesize();
1539 size = ROUND_UP(bytes, pagesize);
1541 #if !defined(ALWAYS_PIC) && defined(x86_64_HOST_ARCH)
1544 if (mmap_32bit_base != 0) {
1545 map_addr = mmap_32bit_base;
1549 result = mmap(map_addr, size, PROT_EXEC|PROT_READ|PROT_WRITE,
1550 MAP_PRIVATE|TRY_MAP_32BIT|fixed|flags, fd, 0);
1552 if (result == MAP_FAILED) {
1553 sysErrorBelch("mmap %lu bytes at %p",(lnat)size,map_addr);
1554 errorBelch("Try specifying an address with +RTS -xm<addr> -RTS");
1555 stg_exit(EXIT_FAILURE);
1558 #if !defined(ALWAYS_PIC) && defined(x86_64_HOST_ARCH)
1559 if (mmap_32bit_base != 0) {
1560 if (result == map_addr) {
1561 mmap_32bit_base = (StgWord8*)map_addr + size;
1563 if ((W_)result > 0x80000000) {
1564 // oops, we were given memory over 2Gb
1565 #if defined(freebsd_HOST_OS) || defined(dragonfly_HOST_OS)
1566 // Some platforms require MAP_FIXED. This is normally
1567 // a bad idea, because MAP_FIXED will overwrite
1568 // existing mappings.
1569 munmap(result,size);
1573 barf("loadObj: failed to mmap() memory below 2Gb; asked for %lu bytes at %p. Try specifying an address with +RTS -xm<addr> -RTS", size, map_addr, result);
1576 // hmm, we were given memory somewhere else, but it's
1577 // still under 2Gb so we can use it. Next time, ask
1578 // for memory right after the place we just got some
1579 mmap_32bit_base = (StgWord8*)result + size;
1583 if ((W_)result > 0x80000000) {
1584 // oops, we were given memory over 2Gb
1585 // ... try allocating memory somewhere else?;
1586 debugTrace(DEBUG_linker,"MAP_32BIT didn't work; gave us %lu bytes at 0x%p", bytes, result);
1587 munmap(result, size);
1589 // Set a base address and try again... (guess: 1Gb)
1590 mmap_32bit_base = (void*)0x40000000;
1600 /* -----------------------------------------------------------------------------
1601 * Load an obj (populate the global symbol table, but don't resolve yet)
1603 * Returns: 1 if ok, 0 on error.
1606 loadObj( char *path )
1616 IF_DEBUG(linker, debugBelch("loadObj %s\n", path));
1619 /* debugBelch("loadObj %s\n", path ); */
1621 /* Check that we haven't already loaded this object.
1622 Ignore requests to load multiple times */
1626 for (o = objects; o; o = o->next) {
1627 if (0 == strcmp(o->fileName, path)) {
1629 break; /* don't need to search further */
1633 IF_DEBUG(linker, debugBelch(
1634 "GHCi runtime linker: warning: looks like you're trying to load the\n"
1635 "same object file twice:\n"
1637 "GHCi will ignore this, but be warned.\n"
1639 return 1; /* success */
1643 oc = stgMallocBytes(sizeof(ObjectCode), "loadObj(oc)");
1645 # if defined(OBJFORMAT_ELF)
1646 oc->formatName = "ELF";
1647 # elif defined(OBJFORMAT_PEi386)
1648 oc->formatName = "PEi386";
1649 # elif defined(OBJFORMAT_MACHO)
1650 oc->formatName = "Mach-O";
1653 barf("loadObj: not implemented on this platform");
1656 r = stat(path, &st);
1658 IF_DEBUG(linker, debugBelch("File doesn't exist\n"));
1662 /* sigh, strdup() isn't a POSIX function, so do it the long way */
1663 oc->fileName = stgMallocBytes( strlen(path)+1, "loadObj" );
1664 strcpy(oc->fileName, path);
1666 oc->fileSize = st.st_size;
1668 oc->sections = NULL;
1669 oc->proddables = NULL;
1671 /* chain it onto the list of objects */
1676 /* On many architectures malloc'd memory isn't executable, so we need to use mmap. */
1678 #if defined(openbsd_HOST_OS)
1679 fd = open(path, O_RDONLY, S_IRUSR);
1681 fd = open(path, O_RDONLY);
1684 barf("loadObj: can't open `%s'", path);
1686 oc->image = mmapForLinker(oc->fileSize, 0, fd);
1690 #else /* !USE_MMAP */
1691 /* load the image into memory */
1692 f = fopen(path, "rb");
1694 barf("loadObj: can't read `%s'", path);
1696 # if defined(mingw32_HOST_OS)
1697 // TODO: We would like to use allocateExec here, but allocateExec
1698 // cannot currently allocate blocks large enough.
1699 oc->image = VirtualAlloc(NULL, oc->fileSize, MEM_RESERVE | MEM_COMMIT,
1700 PAGE_EXECUTE_READWRITE);
1701 # elif defined(darwin_HOST_OS)
1702 // In a Mach-O .o file, all sections can and will be misaligned
1703 // if the total size of the headers is not a multiple of the
1704 // desired alignment. This is fine for .o files that only serve
1705 // as input for the static linker, but it's not fine for us,
1706 // as SSE (used by gcc for floating point) and Altivec require
1707 // 16-byte alignment.
1708 // We calculate the correct alignment from the header before
1709 // reading the file, and then we misalign oc->image on purpose so
1710 // that the actual sections end up aligned again.
1711 oc->misalignment = machoGetMisalignment(f);
1712 oc->image = stgMallocBytes(oc->fileSize + oc->misalignment, "loadObj(image)");
1713 oc->image += oc->misalignment;
1715 oc->image = stgMallocBytes(oc->fileSize, "loadObj(image)");
1720 n = fread ( oc->image, 1, oc->fileSize, f );
1721 if (n != oc->fileSize)
1722 barf("loadObj: error whilst reading `%s'", path);
1725 #endif /* USE_MMAP */
1727 # if defined(OBJFORMAT_MACHO) && (defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH))
1728 r = ocAllocateSymbolExtras_MachO ( oc );
1730 IF_DEBUG(linker, debugBelch("ocAllocateSymbolExtras_MachO failed\n"));
1733 # elif defined(OBJFORMAT_ELF) && (defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH))
1734 r = ocAllocateSymbolExtras_ELF ( oc );
1736 IF_DEBUG(linker, debugBelch("ocAllocateSymbolExtras_ELF failed\n"));
1741 /* verify the in-memory image */
1742 # if defined(OBJFORMAT_ELF)
1743 r = ocVerifyImage_ELF ( oc );
1744 # elif defined(OBJFORMAT_PEi386)
1745 r = ocVerifyImage_PEi386 ( oc );
1746 # elif defined(OBJFORMAT_MACHO)
1747 r = ocVerifyImage_MachO ( oc );
1749 barf("loadObj: no verify method");
1752 IF_DEBUG(linker, debugBelch("ocVerifyImage_* failed\n"));
1756 /* build the symbol list for this image */
1757 # if defined(OBJFORMAT_ELF)
1758 r = ocGetNames_ELF ( oc );
1759 # elif defined(OBJFORMAT_PEi386)
1760 r = ocGetNames_PEi386 ( oc );
1761 # elif defined(OBJFORMAT_MACHO)
1762 r = ocGetNames_MachO ( oc );
1764 barf("loadObj: no getNames method");
1767 IF_DEBUG(linker, debugBelch("ocGetNames_* failed\n"));
1771 /* loaded, but not resolved yet */
1772 oc->status = OBJECT_LOADED;
1777 /* -----------------------------------------------------------------------------
1778 * resolve all the currently unlinked objects in memory
1780 * Returns: 1 if ok, 0 on error.
1790 for (oc = objects; oc; oc = oc->next) {
1791 if (oc->status != OBJECT_RESOLVED) {
1792 # if defined(OBJFORMAT_ELF)
1793 r = ocResolve_ELF ( oc );
1794 # elif defined(OBJFORMAT_PEi386)
1795 r = ocResolve_PEi386 ( oc );
1796 # elif defined(OBJFORMAT_MACHO)
1797 r = ocResolve_MachO ( oc );
1799 barf("resolveObjs: not implemented on this platform");
1801 if (!r) { return r; }
1802 oc->status = OBJECT_RESOLVED;
1808 /* -----------------------------------------------------------------------------
1809 * delete an object from the pool
1812 unloadObj( char *path )
1814 ObjectCode *oc, *prev;
1816 ASSERT(symhash != NULL);
1817 ASSERT(objects != NULL);
1822 for (oc = objects; oc; prev = oc, oc = oc->next) {
1823 if (!strcmp(oc->fileName,path)) {
1825 /* Remove all the mappings for the symbols within this
1830 for (i = 0; i < oc->n_symbols; i++) {
1831 if (oc->symbols[i] != NULL) {
1832 removeStrHashTable(symhash, oc->symbols[i], NULL);
1840 prev->next = oc->next;
1843 // We're going to leave this in place, in case there are
1844 // any pointers from the heap into it:
1845 // #ifdef mingw32_HOST_OS
1846 // VirtualFree(oc->image);
1848 // stgFree(oc->image);
1850 stgFree(oc->fileName);
1851 stgFree(oc->symbols);
1852 stgFree(oc->sections);
1858 errorBelch("unloadObj: can't find `%s' to unload", path);
1862 /* -----------------------------------------------------------------------------
1863 * Sanity checking. For each ObjectCode, maintain a list of address ranges
1864 * which may be prodded during relocation, and abort if we try and write
1865 * outside any of these.
1867 static void addProddableBlock ( ObjectCode* oc, void* start, int size )
1870 = stgMallocBytes(sizeof(ProddableBlock), "addProddableBlock");
1871 /* debugBelch("aPB %p %p %d\n", oc, start, size); */
1875 pb->next = oc->proddables;
1876 oc->proddables = pb;
1879 static void checkProddableBlock ( ObjectCode* oc, void* addr )
1882 for (pb = oc->proddables; pb != NULL; pb = pb->next) {
1883 char* s = (char*)(pb->start);
1884 char* e = s + pb->size - 1;
1885 char* a = (char*)addr;
1886 /* Assumes that the biggest fixup involves a 4-byte write. This
1887 probably needs to be changed to 8 (ie, +7) on 64-bit
1889 if (a >= s && (a+3) <= e) return;
1891 barf("checkProddableBlock: invalid fixup in runtime linker");
1894 /* -----------------------------------------------------------------------------
1895 * Section management.
1897 static void addSection ( ObjectCode* oc, SectionKind kind,
1898 void* start, void* end )
1900 Section* s = stgMallocBytes(sizeof(Section), "addSection");
1904 s->next = oc->sections;
1907 debugBelch("addSection: %p-%p (size %d), kind %d\n",
1908 start, ((char*)end)-1, end - start + 1, kind );
1913 /* --------------------------------------------------------------------------
1915 * This is about allocating a small chunk of memory for every symbol in the
1916 * object file. We make sure that the SymboLExtras are always "in range" of
1917 * limited-range PC-relative instructions on various platforms by allocating
1918 * them right next to the object code itself.
1921 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
1924 ocAllocateSymbolExtras
1926 Allocate additional space at the end of the object file image to make room
1927 for jump islands (powerpc, x86_64) and GOT entries (x86_64).
1929 PowerPC relative branch instructions have a 24 bit displacement field.
1930 As PPC code is always 4-byte-aligned, this yields a +-32MB range.
1931 If a particular imported symbol is outside this range, we have to redirect
1932 the jump to a short piece of new code that just loads the 32bit absolute
1933 address and jumps there.
1934 On x86_64, PC-relative jumps and PC-relative accesses to the GOT are limited
1937 This function just allocates space for one SymbolExtra for every
1938 undefined symbol in the object file. The code for the jump islands is
1939 filled in by makeSymbolExtra below.
1942 static int ocAllocateSymbolExtras( ObjectCode* oc, int count, int first )
1949 int misalignment = 0;
1950 #ifdef darwin_HOST_OS
1951 misalignment = oc->misalignment;
1957 // round up to the nearest 4
1958 aligned = (oc->fileSize + 3) & ~3;
1961 pagesize = getpagesize();
1962 n = ROUND_UP( oc->fileSize, pagesize );
1963 m = ROUND_UP( aligned + sizeof (SymbolExtra) * count, pagesize );
1965 /* we try to use spare space at the end of the last page of the
1966 * image for the jump islands, but if there isn't enough space
1967 * then we have to map some (anonymously, remembering MAP_32BIT).
1969 if( m > n ) // we need to allocate more pages
1971 oc->symbol_extras = mmapForLinker(sizeof(SymbolExtra) * count,
1976 oc->symbol_extras = (SymbolExtra *) (oc->image + aligned);
1979 oc->image -= misalignment;
1980 oc->image = stgReallocBytes( oc->image,
1982 aligned + sizeof (SymbolExtra) * count,
1983 "ocAllocateSymbolExtras" );
1984 oc->image += misalignment;
1986 oc->symbol_extras = (SymbolExtra *) (oc->image + aligned);
1987 #endif /* USE_MMAP */
1989 memset( oc->symbol_extras, 0, sizeof (SymbolExtra) * count );
1992 oc->symbol_extras = NULL;
1994 oc->first_symbol_extra = first;
1995 oc->n_symbol_extras = count;
2000 static SymbolExtra* makeSymbolExtra( ObjectCode* oc,
2001 unsigned long symbolNumber,
2002 unsigned long target )
2006 ASSERT( symbolNumber >= oc->first_symbol_extra
2007 && symbolNumber - oc->first_symbol_extra < oc->n_symbol_extras);
2009 extra = &oc->symbol_extras[symbolNumber - oc->first_symbol_extra];
2011 #ifdef powerpc_HOST_ARCH
2012 // lis r12, hi16(target)
2013 extra->jumpIsland.lis_r12 = 0x3d80;
2014 extra->jumpIsland.hi_addr = target >> 16;
2016 // ori r12, r12, lo16(target)
2017 extra->jumpIsland.ori_r12_r12 = 0x618c;
2018 extra->jumpIsland.lo_addr = target & 0xffff;
2021 extra->jumpIsland.mtctr_r12 = 0x7d8903a6;
2024 extra->jumpIsland.bctr = 0x4e800420;
2026 #ifdef x86_64_HOST_ARCH
2028 static uint8_t jmp[] = { 0xFF, 0x25, 0xF2, 0xFF, 0xFF, 0xFF };
2029 extra->addr = target;
2030 memcpy(extra->jumpIsland, jmp, 6);
2038 /* --------------------------------------------------------------------------
2039 * PowerPC specifics (instruction cache flushing)
2040 * ------------------------------------------------------------------------*/
2042 #ifdef powerpc_HOST_ARCH
2044 ocFlushInstructionCache
2046 Flush the data & instruction caches.
2047 Because the PPC has split data/instruction caches, we have to
2048 do that whenever we modify code at runtime.
2051 static void ocFlushInstructionCache( ObjectCode *oc )
2053 int n = (oc->fileSize + sizeof( SymbolExtra ) * oc->n_symbol_extras + 3) / 4;
2054 unsigned long *p = (unsigned long *) oc->image;
2058 __asm__ volatile ( "dcbf 0,%0\n\t"
2066 __asm__ volatile ( "sync\n\t"
2072 /* --------------------------------------------------------------------------
2073 * PEi386 specifics (Win32 targets)
2074 * ------------------------------------------------------------------------*/
2076 /* The information for this linker comes from
2077 Microsoft Portable Executable
2078 and Common Object File Format Specification
2079 revision 5.1 January 1998
2080 which SimonM says comes from the MS Developer Network CDs.
2082 It can be found there (on older CDs), but can also be found
2085 http://www.microsoft.com/hwdev/hardware/PECOFF.asp
2087 (this is Rev 6.0 from February 1999).
2089 Things move, so if that fails, try searching for it via
2091 http://www.google.com/search?q=PE+COFF+specification
2093 The ultimate reference for the PE format is the Winnt.h
2094 header file that comes with the Platform SDKs; as always,
2095 implementations will drift wrt their documentation.
2097 A good background article on the PE format is Matt Pietrek's
2098 March 1994 article in Microsoft System Journal (MSJ)
2099 (Vol.9, No. 3): "Peering Inside the PE: A Tour of the
2100 Win32 Portable Executable File Format." The info in there
2101 has recently been updated in a two part article in
2102 MSDN magazine, issues Feb and March 2002,
2103 "Inside Windows: An In-Depth Look into the Win32 Portable
2104 Executable File Format"
2106 John Levine's book "Linkers and Loaders" contains useful
2111 #if defined(OBJFORMAT_PEi386)
2115 typedef unsigned char UChar;
2116 typedef unsigned short UInt16;
2117 typedef unsigned int UInt32;
2124 UInt16 NumberOfSections;
2125 UInt32 TimeDateStamp;
2126 UInt32 PointerToSymbolTable;
2127 UInt32 NumberOfSymbols;
2128 UInt16 SizeOfOptionalHeader;
2129 UInt16 Characteristics;
2133 #define sizeof_COFF_header 20
2140 UInt32 VirtualAddress;
2141 UInt32 SizeOfRawData;
2142 UInt32 PointerToRawData;
2143 UInt32 PointerToRelocations;
2144 UInt32 PointerToLinenumbers;
2145 UInt16 NumberOfRelocations;
2146 UInt16 NumberOfLineNumbers;
2147 UInt32 Characteristics;
2151 #define sizeof_COFF_section 40
2158 UInt16 SectionNumber;
2161 UChar NumberOfAuxSymbols;
2165 #define sizeof_COFF_symbol 18
2170 UInt32 VirtualAddress;
2171 UInt32 SymbolTableIndex;
2176 #define sizeof_COFF_reloc 10
2179 /* From PE spec doc, section 3.3.2 */
2180 /* Note use of MYIMAGE_* since IMAGE_* are already defined in
2181 windows.h -- for the same purpose, but I want to know what I'm
2183 #define MYIMAGE_FILE_RELOCS_STRIPPED 0x0001
2184 #define MYIMAGE_FILE_EXECUTABLE_IMAGE 0x0002
2185 #define MYIMAGE_FILE_DLL 0x2000
2186 #define MYIMAGE_FILE_SYSTEM 0x1000
2187 #define MYIMAGE_FILE_BYTES_REVERSED_HI 0x8000
2188 #define MYIMAGE_FILE_BYTES_REVERSED_LO 0x0080
2189 #define MYIMAGE_FILE_32BIT_MACHINE 0x0100
2191 /* From PE spec doc, section 5.4.2 and 5.4.4 */
2192 #define MYIMAGE_SYM_CLASS_EXTERNAL 2
2193 #define MYIMAGE_SYM_CLASS_STATIC 3
2194 #define MYIMAGE_SYM_UNDEFINED 0
2196 /* From PE spec doc, section 4.1 */
2197 #define MYIMAGE_SCN_CNT_CODE 0x00000020
2198 #define MYIMAGE_SCN_CNT_INITIALIZED_DATA 0x00000040
2199 #define MYIMAGE_SCN_LNK_NRELOC_OVFL 0x01000000
2201 /* From PE spec doc, section 5.2.1 */
2202 #define MYIMAGE_REL_I386_DIR32 0x0006
2203 #define MYIMAGE_REL_I386_REL32 0x0014
2206 /* We use myindex to calculate array addresses, rather than
2207 simply doing the normal subscript thing. That's because
2208 some of the above structs have sizes which are not
2209 a whole number of words. GCC rounds their sizes up to a
2210 whole number of words, which means that the address calcs
2211 arising from using normal C indexing or pointer arithmetic
2212 are just plain wrong. Sigh.
2215 myindex ( int scale, void* base, int index )
2218 ((UChar*)base) + scale * index;
2223 printName ( UChar* name, UChar* strtab )
2225 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2226 UInt32 strtab_offset = * (UInt32*)(name+4);
2227 debugBelch("%s", strtab + strtab_offset );
2230 for (i = 0; i < 8; i++) {
2231 if (name[i] == 0) break;
2232 debugBelch("%c", name[i] );
2239 copyName ( UChar* name, UChar* strtab, UChar* dst, int dstSize )
2241 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2242 UInt32 strtab_offset = * (UInt32*)(name+4);
2243 strncpy ( (char*)dst, (char*)strtab+strtab_offset, dstSize );
2249 if (name[i] == 0) break;
2259 cstring_from_COFF_symbol_name ( UChar* name, UChar* strtab )
2262 /* If the string is longer than 8 bytes, look in the
2263 string table for it -- this will be correctly zero terminated.
2265 if (name[0]==0 && name[1]==0 && name[2]==0 && name[3]==0) {
2266 UInt32 strtab_offset = * (UInt32*)(name+4);
2267 return ((UChar*)strtab) + strtab_offset;
2269 /* Otherwise, if shorter than 8 bytes, return the original,
2270 which by defn is correctly terminated.
2272 if (name[7]==0) return name;
2273 /* The annoying case: 8 bytes. Copy into a temporary
2274 (which is never freed ...)
2276 newstr = stgMallocBytes(9, "cstring_from_COFF_symbol_name");
2278 strncpy((char*)newstr,(char*)name,8);
2284 /* Just compares the short names (first 8 chars) */
2285 static COFF_section *
2286 findPEi386SectionCalled ( ObjectCode* oc, UChar* name )
2290 = (COFF_header*)(oc->image);
2291 COFF_section* sectab
2293 ((UChar*)(oc->image))
2294 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2296 for (i = 0; i < hdr->NumberOfSections; i++) {
2299 COFF_section* section_i
2301 myindex ( sizeof_COFF_section, sectab, i );
2302 n1 = (UChar*) &(section_i->Name);
2304 if (n1[0]==n2[0] && n1[1]==n2[1] && n1[2]==n2[2] &&
2305 n1[3]==n2[3] && n1[4]==n2[4] && n1[5]==n2[5] &&
2306 n1[6]==n2[6] && n1[7]==n2[7])
2315 zapTrailingAtSign ( UChar* sym )
2317 # define my_isdigit(c) ((c) >= '0' && (c) <= '9')
2319 if (sym[0] == 0) return;
2321 while (sym[i] != 0) i++;
2324 while (j > 0 && my_isdigit(sym[j])) j--;
2325 if (j > 0 && sym[j] == '@' && j != i) sym[j] = 0;
2330 lookupSymbolInDLLs ( UChar *lbl )
2335 for (o_dll = opened_dlls; o_dll != NULL; o_dll = o_dll->next) {
2336 /* debugBelch("look in %s for %s\n", o_dll->name, lbl); */
2338 if (lbl[0] == '_') {
2339 /* HACK: if the name has an initial underscore, try stripping
2340 it off & look that up first. I've yet to verify whether there's
2341 a Rule that governs whether an initial '_' *should always* be
2342 stripped off when mapping from import lib name to the DLL name.
2344 sym = GetProcAddress(o_dll->instance, (char*)(lbl+1));
2346 /*debugBelch("found %s in %s\n", lbl+1,o_dll->name);*/
2350 sym = GetProcAddress(o_dll->instance, (char*)lbl);
2352 /*debugBelch("found %s in %s\n", lbl,o_dll->name);*/
2361 ocVerifyImage_PEi386 ( ObjectCode* oc )
2366 COFF_section* sectab;
2367 COFF_symbol* symtab;
2369 /* debugBelch("\nLOADING %s\n", oc->fileName); */
2370 hdr = (COFF_header*)(oc->image);
2371 sectab = (COFF_section*) (
2372 ((UChar*)(oc->image))
2373 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2375 symtab = (COFF_symbol*) (
2376 ((UChar*)(oc->image))
2377 + hdr->PointerToSymbolTable
2379 strtab = ((UChar*)symtab)
2380 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2382 if (hdr->Machine != 0x14c) {
2383 errorBelch("%s: Not x86 PEi386", oc->fileName);
2386 if (hdr->SizeOfOptionalHeader != 0) {
2387 errorBelch("%s: PEi386 with nonempty optional header", oc->fileName);
2390 if ( /* (hdr->Characteristics & MYIMAGE_FILE_RELOCS_STRIPPED) || */
2391 (hdr->Characteristics & MYIMAGE_FILE_EXECUTABLE_IMAGE) ||
2392 (hdr->Characteristics & MYIMAGE_FILE_DLL) ||
2393 (hdr->Characteristics & MYIMAGE_FILE_SYSTEM) ) {
2394 errorBelch("%s: Not a PEi386 object file", oc->fileName);
2397 if ( (hdr->Characteristics & MYIMAGE_FILE_BYTES_REVERSED_HI)
2398 /* || !(hdr->Characteristics & MYIMAGE_FILE_32BIT_MACHINE) */ ) {
2399 errorBelch("%s: Invalid PEi386 word size or endiannness: %d",
2401 (int)(hdr->Characteristics));
2404 /* If the string table size is way crazy, this might indicate that
2405 there are more than 64k relocations, despite claims to the
2406 contrary. Hence this test. */
2407 /* debugBelch("strtab size %d\n", * (UInt32*)strtab); */
2409 if ( (*(UInt32*)strtab) > 600000 ) {
2410 /* Note that 600k has no special significance other than being
2411 big enough to handle the almost-2MB-sized lumps that
2412 constitute HSwin32*.o. */
2413 debugBelch("PEi386 object has suspiciously large string table; > 64k relocs?");
2418 /* No further verification after this point; only debug printing. */
2420 IF_DEBUG(linker, i=1);
2421 if (i == 0) return 1;
2423 debugBelch( "sectab offset = %d\n", ((UChar*)sectab) - ((UChar*)hdr) );
2424 debugBelch( "symtab offset = %d\n", ((UChar*)symtab) - ((UChar*)hdr) );
2425 debugBelch( "strtab offset = %d\n", ((UChar*)strtab) - ((UChar*)hdr) );
2428 debugBelch( "Machine: 0x%x\n", (UInt32)(hdr->Machine) );
2429 debugBelch( "# sections: %d\n", (UInt32)(hdr->NumberOfSections) );
2430 debugBelch( "time/date: 0x%x\n", (UInt32)(hdr->TimeDateStamp) );
2431 debugBelch( "symtab offset: %d\n", (UInt32)(hdr->PointerToSymbolTable) );
2432 debugBelch( "# symbols: %d\n", (UInt32)(hdr->NumberOfSymbols) );
2433 debugBelch( "sz of opt hdr: %d\n", (UInt32)(hdr->SizeOfOptionalHeader) );
2434 debugBelch( "characteristics: 0x%x\n", (UInt32)(hdr->Characteristics) );
2436 /* Print the section table. */
2438 for (i = 0; i < hdr->NumberOfSections; i++) {
2440 COFF_section* sectab_i
2442 myindex ( sizeof_COFF_section, sectab, i );
2449 printName ( sectab_i->Name, strtab );
2459 sectab_i->VirtualSize,
2460 sectab_i->VirtualAddress,
2461 sectab_i->SizeOfRawData,
2462 sectab_i->PointerToRawData,
2463 sectab_i->NumberOfRelocations,
2464 sectab_i->PointerToRelocations,
2465 sectab_i->PointerToRawData
2467 reltab = (COFF_reloc*) (
2468 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
2471 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
2472 /* If the relocation field (a short) has overflowed, the
2473 * real count can be found in the first reloc entry.
2475 * See Section 4.1 (last para) of the PE spec (rev6.0).
2477 COFF_reloc* rel = (COFF_reloc*)
2478 myindex ( sizeof_COFF_reloc, reltab, 0 );
2479 noRelocs = rel->VirtualAddress;
2482 noRelocs = sectab_i->NumberOfRelocations;
2486 for (; j < noRelocs; j++) {
2488 COFF_reloc* rel = (COFF_reloc*)
2489 myindex ( sizeof_COFF_reloc, reltab, j );
2491 " type 0x%-4x vaddr 0x%-8x name `",
2493 rel->VirtualAddress );
2494 sym = (COFF_symbol*)
2495 myindex ( sizeof_COFF_symbol, symtab, rel->SymbolTableIndex );
2496 /* Hmm..mysterious looking offset - what's it for? SOF */
2497 printName ( sym->Name, strtab -10 );
2504 debugBelch("string table has size 0x%x\n", * (UInt32*)strtab );
2505 debugBelch("---START of string table---\n");
2506 for (i = 4; i < *(Int32*)strtab; i++) {
2508 debugBelch("\n"); else
2509 debugBelch("%c", strtab[i] );
2511 debugBelch("--- END of string table---\n");
2516 COFF_symbol* symtab_i;
2517 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2518 symtab_i = (COFF_symbol*)
2519 myindex ( sizeof_COFF_symbol, symtab, i );
2525 printName ( symtab_i->Name, strtab );
2534 (Int32)(symtab_i->SectionNumber),
2535 (UInt32)symtab_i->Type,
2536 (UInt32)symtab_i->StorageClass,
2537 (UInt32)symtab_i->NumberOfAuxSymbols
2539 i += symtab_i->NumberOfAuxSymbols;
2549 ocGetNames_PEi386 ( ObjectCode* oc )
2552 COFF_section* sectab;
2553 COFF_symbol* symtab;
2560 hdr = (COFF_header*)(oc->image);
2561 sectab = (COFF_section*) (
2562 ((UChar*)(oc->image))
2563 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2565 symtab = (COFF_symbol*) (
2566 ((UChar*)(oc->image))
2567 + hdr->PointerToSymbolTable
2569 strtab = ((UChar*)(oc->image))
2570 + hdr->PointerToSymbolTable
2571 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2573 /* Allocate space for any (local, anonymous) .bss sections. */
2575 for (i = 0; i < hdr->NumberOfSections; i++) {
2578 COFF_section* sectab_i
2580 myindex ( sizeof_COFF_section, sectab, i );
2581 if (0 != strcmp((char*)sectab_i->Name, ".bss")) continue;
2582 /* sof 10/05: the PE spec text isn't too clear regarding what
2583 * the SizeOfRawData field is supposed to hold for object
2584 * file sections containing just uninitialized data -- for executables,
2585 * it is supposed to be zero; unclear what it's supposed to be
2586 * for object files. However, VirtualSize is guaranteed to be
2587 * zero for object files, which definitely suggests that SizeOfRawData
2588 * will be non-zero (where else would the size of this .bss section be
2589 * stored?) Looking at the COFF_section info for incoming object files,
2590 * this certainly appears to be the case.
2592 * => I suspect we've been incorrectly handling .bss sections in (relocatable)
2593 * object files up until now. This turned out to bite us with ghc-6.4.1's use
2594 * of gcc-3.4.x, which has started to emit initially-zeroed-out local 'static'
2595 * variable decls into to the .bss section. (The specific function in Q which
2596 * triggered this is libraries/base/cbits/dirUtils.c:__hscore_getFolderPath())
2598 if (sectab_i->VirtualSize == 0 && sectab_i->SizeOfRawData == 0) continue;
2599 /* This is a non-empty .bss section. Allocate zeroed space for
2600 it, and set its PointerToRawData field such that oc->image +
2601 PointerToRawData == addr_of_zeroed_space. */
2602 bss_sz = sectab_i->VirtualSize;
2603 if ( bss_sz < sectab_i->SizeOfRawData) { bss_sz = sectab_i->SizeOfRawData; }
2604 zspace = stgCallocBytes(1, bss_sz, "ocGetNames_PEi386(anonymous bss)");
2605 sectab_i->PointerToRawData = ((UChar*)zspace) - ((UChar*)(oc->image));
2606 addProddableBlock(oc, zspace, bss_sz);
2607 /* debugBelch("BSS anon section at 0x%x\n", zspace); */
2610 /* Copy section information into the ObjectCode. */
2612 for (i = 0; i < hdr->NumberOfSections; i++) {
2618 = SECTIONKIND_OTHER;
2619 COFF_section* sectab_i
2621 myindex ( sizeof_COFF_section, sectab, i );
2622 IF_DEBUG(linker, debugBelch("section name = %s\n", sectab_i->Name ));
2625 /* I'm sure this is the Right Way to do it. However, the
2626 alternative of testing the sectab_i->Name field seems to
2627 work ok with Cygwin.
2629 if (sectab_i->Characteristics & MYIMAGE_SCN_CNT_CODE ||
2630 sectab_i->Characteristics & MYIMAGE_SCN_CNT_INITIALIZED_DATA)
2631 kind = SECTIONKIND_CODE_OR_RODATA;
2634 if (0==strcmp(".text",(char*)sectab_i->Name) ||
2635 0==strcmp(".rdata",(char*)sectab_i->Name)||
2636 0==strcmp(".rodata",(char*)sectab_i->Name))
2637 kind = SECTIONKIND_CODE_OR_RODATA;
2638 if (0==strcmp(".data",(char*)sectab_i->Name) ||
2639 0==strcmp(".bss",(char*)sectab_i->Name))
2640 kind = SECTIONKIND_RWDATA;
2642 ASSERT(sectab_i->SizeOfRawData == 0 || sectab_i->VirtualSize == 0);
2643 sz = sectab_i->SizeOfRawData;
2644 if (sz < sectab_i->VirtualSize) sz = sectab_i->VirtualSize;
2646 start = ((UChar*)(oc->image)) + sectab_i->PointerToRawData;
2647 end = start + sz - 1;
2649 if (kind == SECTIONKIND_OTHER
2650 /* Ignore sections called which contain stabs debugging
2652 && 0 != strcmp(".stab", (char*)sectab_i->Name)
2653 && 0 != strcmp(".stabstr", (char*)sectab_i->Name)
2654 /* ignore constructor section for now */
2655 && 0 != strcmp(".ctors", (char*)sectab_i->Name)
2656 /* ignore section generated from .ident */
2657 && 0!= strcmp("/4", (char*)sectab_i->Name)
2658 /* ignore unknown section that appeared in gcc 3.4.5(?) */
2659 && 0!= strcmp(".reloc", (char*)sectab_i->Name)
2661 errorBelch("Unknown PEi386 section name `%s' (while processing: %s)", sectab_i->Name, oc->fileName);
2665 if (kind != SECTIONKIND_OTHER && end >= start) {
2666 addSection(oc, kind, start, end);
2667 addProddableBlock(oc, start, end - start + 1);
2671 /* Copy exported symbols into the ObjectCode. */
2673 oc->n_symbols = hdr->NumberOfSymbols;
2674 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
2675 "ocGetNames_PEi386(oc->symbols)");
2676 /* Call me paranoid; I don't care. */
2677 for (i = 0; i < oc->n_symbols; i++)
2678 oc->symbols[i] = NULL;
2682 COFF_symbol* symtab_i;
2683 if (i >= (Int32)(hdr->NumberOfSymbols)) break;
2684 symtab_i = (COFF_symbol*)
2685 myindex ( sizeof_COFF_symbol, symtab, i );
2689 if (symtab_i->StorageClass == MYIMAGE_SYM_CLASS_EXTERNAL
2690 && symtab_i->SectionNumber != MYIMAGE_SYM_UNDEFINED) {
2691 /* This symbol is global and defined, viz, exported */
2692 /* for MYIMAGE_SYMCLASS_EXTERNAL
2693 && !MYIMAGE_SYM_UNDEFINED,
2694 the address of the symbol is:
2695 address of relevant section + offset in section
2697 COFF_section* sectabent
2698 = (COFF_section*) myindex ( sizeof_COFF_section,
2700 symtab_i->SectionNumber-1 );
2701 addr = ((UChar*)(oc->image))
2702 + (sectabent->PointerToRawData
2706 if (symtab_i->SectionNumber == MYIMAGE_SYM_UNDEFINED
2707 && symtab_i->Value > 0) {
2708 /* This symbol isn't in any section at all, ie, global bss.
2709 Allocate zeroed space for it. */
2710 addr = stgCallocBytes(1, symtab_i->Value,
2711 "ocGetNames_PEi386(non-anonymous bss)");
2712 addSection(oc, SECTIONKIND_RWDATA, addr,
2713 ((UChar*)addr) + symtab_i->Value - 1);
2714 addProddableBlock(oc, addr, symtab_i->Value);
2715 /* debugBelch("BSS section at 0x%x\n", addr); */
2718 if (addr != NULL ) {
2719 sname = cstring_from_COFF_symbol_name ( symtab_i->Name, strtab );
2720 /* debugBelch("addSymbol %p `%s \n", addr,sname); */
2721 IF_DEBUG(linker, debugBelch("addSymbol %p `%s'\n", addr,sname);)
2722 ASSERT(i >= 0 && i < oc->n_symbols);
2723 /* cstring_from_COFF_symbol_name always succeeds. */
2724 oc->symbols[i] = (char*)sname;
2725 ghciInsertStrHashTable(oc->fileName, symhash, (char*)sname, addr);
2729 "IGNORING symbol %d\n"
2733 printName ( symtab_i->Name, strtab );
2742 (Int32)(symtab_i->SectionNumber),
2743 (UInt32)symtab_i->Type,
2744 (UInt32)symtab_i->StorageClass,
2745 (UInt32)symtab_i->NumberOfAuxSymbols
2750 i += symtab_i->NumberOfAuxSymbols;
2759 ocResolve_PEi386 ( ObjectCode* oc )
2762 COFF_section* sectab;
2763 COFF_symbol* symtab;
2773 /* ToDo: should be variable-sized? But is at least safe in the
2774 sense of buffer-overrun-proof. */
2776 /* debugBelch("resolving for %s\n", oc->fileName); */
2778 hdr = (COFF_header*)(oc->image);
2779 sectab = (COFF_section*) (
2780 ((UChar*)(oc->image))
2781 + sizeof_COFF_header + hdr->SizeOfOptionalHeader
2783 symtab = (COFF_symbol*) (
2784 ((UChar*)(oc->image))
2785 + hdr->PointerToSymbolTable
2787 strtab = ((UChar*)(oc->image))
2788 + hdr->PointerToSymbolTable
2789 + hdr->NumberOfSymbols * sizeof_COFF_symbol;
2791 for (i = 0; i < hdr->NumberOfSections; i++) {
2792 COFF_section* sectab_i
2794 myindex ( sizeof_COFF_section, sectab, i );
2797 ((UChar*)(oc->image)) + sectab_i->PointerToRelocations
2800 /* Ignore sections called which contain stabs debugging
2802 if (0 == strcmp(".stab", (char*)sectab_i->Name)
2803 || 0 == strcmp(".stabstr", (char*)sectab_i->Name)
2804 || 0 == strcmp(".ctors", (char*)sectab_i->Name))
2807 if ( sectab_i->Characteristics & MYIMAGE_SCN_LNK_NRELOC_OVFL ) {
2808 /* If the relocation field (a short) has overflowed, the
2809 * real count can be found in the first reloc entry.
2811 * See Section 4.1 (last para) of the PE spec (rev6.0).
2813 * Nov2003 update: the GNU linker still doesn't correctly
2814 * handle the generation of relocatable object files with
2815 * overflown relocations. Hence the output to warn of potential
2818 COFF_reloc* rel = (COFF_reloc*)
2819 myindex ( sizeof_COFF_reloc, reltab, 0 );
2820 noRelocs = rel->VirtualAddress;
2822 /* 10/05: we now assume (and check for) a GNU ld that is capable
2823 * of handling object files with (>2^16) of relocs.
2826 debugBelch("WARNING: Overflown relocation field (# relocs found: %u)\n",
2831 noRelocs = sectab_i->NumberOfRelocations;
2836 for (; j < noRelocs; j++) {
2838 COFF_reloc* reltab_j
2840 myindex ( sizeof_COFF_reloc, reltab, j );
2842 /* the location to patch */
2844 ((UChar*)(oc->image))
2845 + (sectab_i->PointerToRawData
2846 + reltab_j->VirtualAddress
2847 - sectab_i->VirtualAddress )
2849 /* the existing contents of pP */
2851 /* the symbol to connect to */
2852 sym = (COFF_symbol*)
2853 myindex ( sizeof_COFF_symbol,
2854 symtab, reltab_j->SymbolTableIndex );
2857 "reloc sec %2d num %3d: type 0x%-4x "
2858 "vaddr 0x%-8x name `",
2860 (UInt32)reltab_j->Type,
2861 reltab_j->VirtualAddress );
2862 printName ( sym->Name, strtab );
2863 debugBelch("'\n" ));
2865 if (sym->StorageClass == MYIMAGE_SYM_CLASS_STATIC) {
2866 COFF_section* section_sym
2867 = findPEi386SectionCalled ( oc, sym->Name );
2869 errorBelch("%s: can't find section `%s'", oc->fileName, sym->Name);
2872 S = ((UInt32)(oc->image))
2873 + (section_sym->PointerToRawData
2876 copyName ( sym->Name, strtab, symbol, 1000-1 );
2877 S = (UInt32) lookupSymbol( (char*)symbol );
2878 if ((void*)S != NULL) goto foundit;
2879 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
2883 checkProddableBlock(oc, pP);
2884 switch (reltab_j->Type) {
2885 case MYIMAGE_REL_I386_DIR32:
2888 case MYIMAGE_REL_I386_REL32:
2889 /* Tricky. We have to insert a displacement at
2890 pP which, when added to the PC for the _next_
2891 insn, gives the address of the target (S).
2892 Problem is to know the address of the next insn
2893 when we only know pP. We assume that this
2894 literal field is always the last in the insn,
2895 so that the address of the next insn is pP+4
2896 -- hence the constant 4.
2897 Also I don't know if A should be added, but so
2898 far it has always been zero.
2900 SOF 05/2005: 'A' (old contents of *pP) have been observed
2901 to contain values other than zero (the 'wx' object file
2902 that came with wxhaskell-0.9.4; dunno how it was compiled..).
2903 So, add displacement to old value instead of asserting
2904 A to be zero. Fixes wxhaskell-related crashes, and no other
2905 ill effects have been observed.
2907 Update: the reason why we're seeing these more elaborate
2908 relocations is due to a switch in how the NCG compiles SRTs
2909 and offsets to them from info tables. SRTs live in .(ro)data,
2910 while info tables live in .text, causing GAS to emit REL32/DISP32
2911 relocations with non-zero values. Adding the displacement is
2912 the right thing to do.
2914 *pP = S - ((UInt32)pP) - 4 + A;
2917 debugBelch("%s: unhandled PEi386 relocation type %d",
2918 oc->fileName, reltab_j->Type);
2925 IF_DEBUG(linker, debugBelch("completed %s", oc->fileName));
2929 #endif /* defined(OBJFORMAT_PEi386) */
2932 /* --------------------------------------------------------------------------
2934 * ------------------------------------------------------------------------*/
2936 #if defined(OBJFORMAT_ELF)
2941 #if defined(sparc_HOST_ARCH)
2942 # define ELF_TARGET_SPARC /* Used inside <elf.h> */
2943 #elif defined(i386_HOST_ARCH)
2944 # define ELF_TARGET_386 /* Used inside <elf.h> */
2945 #elif defined(x86_64_HOST_ARCH)
2946 # define ELF_TARGET_X64_64
2950 #if !defined(openbsd_HOST_OS)
2953 /* openbsd elf has things in different places, with diff names */
2954 # include <elf_abi.h>
2955 # include <machine/reloc.h>
2956 # define R_386_32 RELOC_32
2957 # define R_386_PC32 RELOC_PC32
2960 /* If elf.h doesn't define it */
2961 # ifndef R_X86_64_PC64
2962 # define R_X86_64_PC64 24
2966 * Define a set of types which can be used for both ELF32 and ELF64
2970 #define ELFCLASS ELFCLASS64
2971 #define Elf_Addr Elf64_Addr
2972 #define Elf_Word Elf64_Word
2973 #define Elf_Sword Elf64_Sword
2974 #define Elf_Ehdr Elf64_Ehdr
2975 #define Elf_Phdr Elf64_Phdr
2976 #define Elf_Shdr Elf64_Shdr
2977 #define Elf_Sym Elf64_Sym
2978 #define Elf_Rel Elf64_Rel
2979 #define Elf_Rela Elf64_Rela
2981 #define ELF_ST_TYPE ELF64_ST_TYPE
2984 #define ELF_ST_BIND ELF64_ST_BIND
2987 #define ELF_R_TYPE ELF64_R_TYPE
2990 #define ELF_R_SYM ELF64_R_SYM
2993 #define ELFCLASS ELFCLASS32
2994 #define Elf_Addr Elf32_Addr
2995 #define Elf_Word Elf32_Word
2996 #define Elf_Sword Elf32_Sword
2997 #define Elf_Ehdr Elf32_Ehdr
2998 #define Elf_Phdr Elf32_Phdr
2999 #define Elf_Shdr Elf32_Shdr
3000 #define Elf_Sym Elf32_Sym
3001 #define Elf_Rel Elf32_Rel
3002 #define Elf_Rela Elf32_Rela
3004 #define ELF_ST_TYPE ELF32_ST_TYPE
3007 #define ELF_ST_BIND ELF32_ST_BIND
3010 #define ELF_R_TYPE ELF32_R_TYPE
3013 #define ELF_R_SYM ELF32_R_SYM
3019 * Functions to allocate entries in dynamic sections. Currently we simply
3020 * preallocate a large number, and we don't check if a entry for the given
3021 * target already exists (a linear search is too slow). Ideally these
3022 * entries would be associated with symbols.
3025 /* These sizes sufficient to load HSbase + HShaskell98 + a few modules */
3026 #define GOT_SIZE 0x20000
3027 #define FUNCTION_TABLE_SIZE 0x10000
3028 #define PLT_SIZE 0x08000
3031 static Elf_Addr got[GOT_SIZE];
3032 static unsigned int gotIndex;
3033 static Elf_Addr gp_val = (Elf_Addr)got;
3036 allocateGOTEntry(Elf_Addr target)
3040 if (gotIndex >= GOT_SIZE)
3041 barf("Global offset table overflow");
3043 entry = &got[gotIndex++];
3045 return (Elf_Addr)entry;
3049 #ifdef ELF_FUNCTION_DESC
3055 static FunctionDesc functionTable[FUNCTION_TABLE_SIZE];
3056 static unsigned int functionTableIndex;
3059 allocateFunctionDesc(Elf_Addr target)
3061 FunctionDesc *entry;
3063 if (functionTableIndex >= FUNCTION_TABLE_SIZE)
3064 barf("Function table overflow");
3066 entry = &functionTable[functionTableIndex++];
3068 entry->gp = (Elf_Addr)gp_val;
3069 return (Elf_Addr)entry;
3073 copyFunctionDesc(Elf_Addr target)
3075 FunctionDesc *olddesc = (FunctionDesc *)target;
3076 FunctionDesc *newdesc;
3078 newdesc = (FunctionDesc *)allocateFunctionDesc(olddesc->ip);
3079 newdesc->gp = olddesc->gp;
3080 return (Elf_Addr)newdesc;
3087 unsigned char code[sizeof(plt_code)];
3091 allocatePLTEntry(Elf_Addr target, ObjectCode *oc)
3093 PLTEntry *plt = (PLTEntry *)oc->plt;
3096 if (oc->pltIndex >= PLT_SIZE)
3097 barf("Procedure table overflow");
3099 entry = &plt[oc->pltIndex++];
3100 memcpy(entry->code, plt_code, sizeof(entry->code));
3101 PLT_RELOC(entry->code, target);
3102 return (Elf_Addr)entry;
3108 return (PLT_SIZE * sizeof(PLTEntry));
3114 * Generic ELF functions
3118 findElfSection ( void* objImage, Elf_Word sh_type )
3120 char* ehdrC = (char*)objImage;
3121 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
3122 Elf_Shdr* shdr = (Elf_Shdr*)(ehdrC + ehdr->e_shoff);
3123 char* sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
3127 for (i = 0; i < ehdr->e_shnum; i++) {
3128 if (shdr[i].sh_type == sh_type
3129 /* Ignore the section header's string table. */
3130 && i != ehdr->e_shstrndx
3131 /* Ignore string tables named .stabstr, as they contain
3133 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
3135 ptr = ehdrC + shdr[i].sh_offset;
3143 ocVerifyImage_ELF ( ObjectCode* oc )
3147 int i, j, nent, nstrtab, nsymtabs;
3151 char* ehdrC = (char*)(oc->image);
3152 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
3154 if (ehdr->e_ident[EI_MAG0] != ELFMAG0 ||
3155 ehdr->e_ident[EI_MAG1] != ELFMAG1 ||
3156 ehdr->e_ident[EI_MAG2] != ELFMAG2 ||
3157 ehdr->e_ident[EI_MAG3] != ELFMAG3) {
3158 errorBelch("%s: not an ELF object", oc->fileName);
3162 if (ehdr->e_ident[EI_CLASS] != ELFCLASS) {
3163 errorBelch("%s: unsupported ELF format", oc->fileName);
3167 if (ehdr->e_ident[EI_DATA] == ELFDATA2LSB) {
3168 IF_DEBUG(linker,debugBelch( "Is little-endian\n" ));
3170 if (ehdr->e_ident[EI_DATA] == ELFDATA2MSB) {
3171 IF_DEBUG(linker,debugBelch( "Is big-endian\n" ));
3173 errorBelch("%s: unknown endiannness", oc->fileName);
3177 if (ehdr->e_type != ET_REL) {
3178 errorBelch("%s: not a relocatable object (.o) file", oc->fileName);
3181 IF_DEBUG(linker, debugBelch( "Is a relocatable object (.o) file\n" ));
3183 IF_DEBUG(linker,debugBelch( "Architecture is " ));
3184 switch (ehdr->e_machine) {
3185 case EM_386: IF_DEBUG(linker,debugBelch( "x86" )); break;
3186 #ifdef EM_SPARC32PLUS
3187 case EM_SPARC32PLUS:
3189 case EM_SPARC: IF_DEBUG(linker,debugBelch( "sparc" )); break;
3191 case EM_IA_64: IF_DEBUG(linker,debugBelch( "ia64" )); break;
3193 case EM_PPC: IF_DEBUG(linker,debugBelch( "powerpc32" )); break;
3195 case EM_X86_64: IF_DEBUG(linker,debugBelch( "x86_64" )); break;
3196 #elif defined(EM_AMD64)
3197 case EM_AMD64: IF_DEBUG(linker,debugBelch( "amd64" )); break;
3199 default: IF_DEBUG(linker,debugBelch( "unknown" ));
3200 errorBelch("%s: unknown architecture (e_machine == %d)"
3201 , oc->fileName, ehdr->e_machine);
3205 IF_DEBUG(linker,debugBelch(
3206 "\nSection header table: start %ld, n_entries %d, ent_size %d\n",
3207 (long)ehdr->e_shoff, ehdr->e_shnum, ehdr->e_shentsize ));
3209 ASSERT (ehdr->e_shentsize == sizeof(Elf_Shdr));
3211 shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3213 if (ehdr->e_shstrndx == SHN_UNDEF) {
3214 errorBelch("%s: no section header string table", oc->fileName);
3217 IF_DEBUG(linker,debugBelch( "Section header string table is section %d\n",
3219 sh_strtab = ehdrC + shdr[ehdr->e_shstrndx].sh_offset;
3222 for (i = 0; i < ehdr->e_shnum; i++) {
3223 IF_DEBUG(linker,debugBelch("%2d: ", i ));
3224 IF_DEBUG(linker,debugBelch("type=%2d ", (int)shdr[i].sh_type ));
3225 IF_DEBUG(linker,debugBelch("size=%4d ", (int)shdr[i].sh_size ));
3226 IF_DEBUG(linker,debugBelch("offs=%4d ", (int)shdr[i].sh_offset ));
3227 IF_DEBUG(linker,debugBelch(" (%p .. %p) ",
3228 ehdrC + shdr[i].sh_offset,
3229 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1));
3231 if (shdr[i].sh_type == SHT_REL) {
3232 IF_DEBUG(linker,debugBelch("Rel " ));
3233 } else if (shdr[i].sh_type == SHT_RELA) {
3234 IF_DEBUG(linker,debugBelch("RelA " ));
3236 IF_DEBUG(linker,debugBelch(" "));
3239 IF_DEBUG(linker,debugBelch("sname=%s\n", sh_strtab + shdr[i].sh_name ));
3243 IF_DEBUG(linker,debugBelch( "\nString tables" ));
3246 for (i = 0; i < ehdr->e_shnum; i++) {
3247 if (shdr[i].sh_type == SHT_STRTAB
3248 /* Ignore the section header's string table. */
3249 && i != ehdr->e_shstrndx
3250 /* Ignore string tables named .stabstr, as they contain
3252 && 0 != memcmp(".stabstr", sh_strtab + shdr[i].sh_name, 8)
3254 IF_DEBUG(linker,debugBelch(" section %d is a normal string table", i ));
3255 strtab = ehdrC + shdr[i].sh_offset;
3260 errorBelch("%s: no string tables, or too many", oc->fileName);
3265 IF_DEBUG(linker,debugBelch( "\nSymbol tables" ));
3266 for (i = 0; i < ehdr->e_shnum; i++) {
3267 if (shdr[i].sh_type != SHT_SYMTAB) continue;
3268 IF_DEBUG(linker,debugBelch( "section %d is a symbol table\n", i ));
3270 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
3271 nent = shdr[i].sh_size / sizeof(Elf_Sym);
3272 IF_DEBUG(linker,debugBelch( " number of entries is apparently %d (%ld rem)\n",
3274 (long)shdr[i].sh_size % sizeof(Elf_Sym)
3276 if (0 != shdr[i].sh_size % sizeof(Elf_Sym)) {
3277 errorBelch("%s: non-integral number of symbol table entries", oc->fileName);
3280 for (j = 0; j < nent; j++) {
3281 IF_DEBUG(linker,debugBelch(" %2d ", j ));
3282 IF_DEBUG(linker,debugBelch(" sec=%-5d size=%-3d val=%5p ",
3283 (int)stab[j].st_shndx,
3284 (int)stab[j].st_size,
3285 (char*)stab[j].st_value ));
3287 IF_DEBUG(linker,debugBelch("type=" ));
3288 switch (ELF_ST_TYPE(stab[j].st_info)) {
3289 case STT_NOTYPE: IF_DEBUG(linker,debugBelch("notype " )); break;
3290 case STT_OBJECT: IF_DEBUG(linker,debugBelch("object " )); break;
3291 case STT_FUNC : IF_DEBUG(linker,debugBelch("func " )); break;
3292 case STT_SECTION: IF_DEBUG(linker,debugBelch("section" )); break;
3293 case STT_FILE: IF_DEBUG(linker,debugBelch("file " )); break;
3294 default: IF_DEBUG(linker,debugBelch("? " )); break;
3296 IF_DEBUG(linker,debugBelch(" " ));
3298 IF_DEBUG(linker,debugBelch("bind=" ));
3299 switch (ELF_ST_BIND(stab[j].st_info)) {
3300 case STB_LOCAL : IF_DEBUG(linker,debugBelch("local " )); break;
3301 case STB_GLOBAL: IF_DEBUG(linker,debugBelch("global" )); break;
3302 case STB_WEAK : IF_DEBUG(linker,debugBelch("weak " )); break;
3303 default: IF_DEBUG(linker,debugBelch("? " )); break;
3305 IF_DEBUG(linker,debugBelch(" " ));
3307 IF_DEBUG(linker,debugBelch("name=%s\n", strtab + stab[j].st_name ));
3311 if (nsymtabs == 0) {
3312 errorBelch("%s: didn't find any symbol tables", oc->fileName);
3319 static int getSectionKind_ELF( Elf_Shdr *hdr, int *is_bss )
3323 if (hdr->sh_type == SHT_PROGBITS
3324 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_EXECINSTR)) {
3325 /* .text-style section */
3326 return SECTIONKIND_CODE_OR_RODATA;
3329 if (hdr->sh_type == SHT_PROGBITS
3330 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
3331 /* .data-style section */
3332 return SECTIONKIND_RWDATA;
3335 if (hdr->sh_type == SHT_PROGBITS
3336 && (hdr->sh_flags & SHF_ALLOC) && !(hdr->sh_flags & SHF_WRITE)) {
3337 /* .rodata-style section */
3338 return SECTIONKIND_CODE_OR_RODATA;
3341 if (hdr->sh_type == SHT_NOBITS
3342 && (hdr->sh_flags & SHF_ALLOC) && (hdr->sh_flags & SHF_WRITE)) {
3343 /* .bss-style section */
3345 return SECTIONKIND_RWDATA;
3348 return SECTIONKIND_OTHER;
3353 ocGetNames_ELF ( ObjectCode* oc )
3358 char* ehdrC = (char*)(oc->image);
3359 Elf_Ehdr* ehdr = (Elf_Ehdr*)ehdrC;
3360 char* strtab = findElfSection ( ehdrC, SHT_STRTAB );
3361 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3363 ASSERT(symhash != NULL);
3366 errorBelch("%s: no strtab", oc->fileName);
3371 for (i = 0; i < ehdr->e_shnum; i++) {
3372 /* Figure out what kind of section it is. Logic derived from
3373 Figure 1.14 ("Special Sections") of the ELF document
3374 ("Portable Formats Specification, Version 1.1"). */
3376 SectionKind kind = getSectionKind_ELF(&shdr[i], &is_bss);
3378 if (is_bss && shdr[i].sh_size > 0) {
3379 /* This is a non-empty .bss section. Allocate zeroed space for
3380 it, and set its .sh_offset field such that
3381 ehdrC + .sh_offset == addr_of_zeroed_space. */
3382 char* zspace = stgCallocBytes(1, shdr[i].sh_size,
3383 "ocGetNames_ELF(BSS)");
3384 shdr[i].sh_offset = ((char*)zspace) - ((char*)ehdrC);
3386 debugBelch("BSS section at 0x%x, size %d\n",
3387 zspace, shdr[i].sh_size);
3391 /* fill in the section info */
3392 if (kind != SECTIONKIND_OTHER && shdr[i].sh_size > 0) {
3393 addProddableBlock(oc, ehdrC + shdr[i].sh_offset, shdr[i].sh_size);
3394 addSection(oc, kind, ehdrC + shdr[i].sh_offset,
3395 ehdrC + shdr[i].sh_offset + shdr[i].sh_size - 1);
3398 if (shdr[i].sh_type != SHT_SYMTAB) continue;
3400 /* copy stuff into this module's object symbol table */
3401 stab = (Elf_Sym*) (ehdrC + shdr[i].sh_offset);
3402 nent = shdr[i].sh_size / sizeof(Elf_Sym);
3404 oc->n_symbols = nent;
3405 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
3406 "ocGetNames_ELF(oc->symbols)");
3408 for (j = 0; j < nent; j++) {
3410 char isLocal = FALSE; /* avoids uninit-var warning */
3412 char* nm = strtab + stab[j].st_name;
3413 int secno = stab[j].st_shndx;
3415 /* Figure out if we want to add it; if so, set ad to its
3416 address. Otherwise leave ad == NULL. */
3418 if (secno == SHN_COMMON) {
3420 ad = stgCallocBytes(1, stab[j].st_size, "ocGetNames_ELF(COMMON)");
3422 debugBelch("COMMON symbol, size %d name %s\n",
3423 stab[j].st_size, nm);
3425 /* Pointless to do addProddableBlock() for this area,
3426 since the linker should never poke around in it. */
3429 if ( ( ELF_ST_BIND(stab[j].st_info)==STB_GLOBAL
3430 || ELF_ST_BIND(stab[j].st_info)==STB_LOCAL
3432 /* and not an undefined symbol */
3433 && stab[j].st_shndx != SHN_UNDEF
3434 /* and not in a "special section" */
3435 && stab[j].st_shndx < SHN_LORESERVE
3437 /* and it's a not a section or string table or anything silly */
3438 ( ELF_ST_TYPE(stab[j].st_info)==STT_FUNC ||
3439 ELF_ST_TYPE(stab[j].st_info)==STT_OBJECT ||
3440 ELF_ST_TYPE(stab[j].st_info)==STT_NOTYPE
3443 /* Section 0 is the undefined section, hence > and not >=. */
3444 ASSERT(secno > 0 && secno < ehdr->e_shnum);
3446 if (shdr[secno].sh_type == SHT_NOBITS) {
3447 debugBelch(" BSS symbol, size %d off %d name %s\n",
3448 stab[j].st_size, stab[j].st_value, nm);
3451 ad = ehdrC + shdr[ secno ].sh_offset + stab[j].st_value;
3452 if (ELF_ST_BIND(stab[j].st_info)==STB_LOCAL) {
3455 #ifdef ELF_FUNCTION_DESC
3456 /* dlsym() and the initialisation table both give us function
3457 * descriptors, so to be consistent we store function descriptors
3458 * in the symbol table */
3459 if (ELF_ST_TYPE(stab[j].st_info) == STT_FUNC)
3460 ad = (char *)allocateFunctionDesc((Elf_Addr)ad);
3462 IF_DEBUG(linker,debugBelch( "addOTabName(GLOB): %10p %s %s\n",
3463 ad, oc->fileName, nm ));
3468 /* And the decision is ... */
3472 oc->symbols[j] = nm;
3475 /* Ignore entirely. */
3477 ghciInsertStrHashTable(oc->fileName, symhash, nm, ad);
3481 IF_DEBUG(linker,debugBelch( "skipping `%s'\n",
3482 strtab + stab[j].st_name ));
3485 "skipping bind = %d, type = %d, shndx = %d `%s'\n",
3486 (int)ELF_ST_BIND(stab[j].st_info),
3487 (int)ELF_ST_TYPE(stab[j].st_info),
3488 (int)stab[j].st_shndx,
3489 strtab + stab[j].st_name
3492 oc->symbols[j] = NULL;
3501 /* Do ELF relocations which lack an explicit addend. All x86-linux
3502 relocations appear to be of this form. */
3504 do_Elf_Rel_relocations ( ObjectCode* oc, char* ehdrC,
3505 Elf_Shdr* shdr, int shnum,
3506 Elf_Sym* stab, char* strtab )
3511 Elf_Rel* rtab = (Elf_Rel*) (ehdrC + shdr[shnum].sh_offset);
3512 int nent = shdr[shnum].sh_size / sizeof(Elf_Rel);
3513 int target_shndx = shdr[shnum].sh_info;
3514 int symtab_shndx = shdr[shnum].sh_link;
3516 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
3517 targ = (Elf_Word*)(ehdrC + shdr[ target_shndx ].sh_offset);
3518 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
3519 target_shndx, symtab_shndx ));
3521 /* Skip sections that we're not interested in. */
3524 SectionKind kind = getSectionKind_ELF(&shdr[target_shndx], &is_bss);
3525 if (kind == SECTIONKIND_OTHER) {
3526 IF_DEBUG(linker,debugBelch( "skipping (target section not loaded)"));
3531 for (j = 0; j < nent; j++) {
3532 Elf_Addr offset = rtab[j].r_offset;
3533 Elf_Addr info = rtab[j].r_info;
3535 Elf_Addr P = ((Elf_Addr)targ) + offset;
3536 Elf_Word* pP = (Elf_Word*)P;
3541 StgStablePtr stablePtr;
3544 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p)",
3545 j, (void*)offset, (void*)info ));
3547 IF_DEBUG(linker,debugBelch( " ZERO" ));
3550 Elf_Sym sym = stab[ELF_R_SYM(info)];
3551 /* First see if it is a local symbol. */
3552 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
3553 /* Yes, so we can get the address directly from the ELF symbol
3555 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
3557 (ehdrC + shdr[ sym.st_shndx ].sh_offset
3558 + stab[ELF_R_SYM(info)].st_value);
3561 symbol = strtab + sym.st_name;
3562 stablePtr = (StgStablePtr)lookupHashTable(stablehash, (StgWord)symbol);
3563 if (NULL == stablePtr) {
3564 /* No, so look up the name in our global table. */
3565 S_tmp = lookupSymbol( symbol );
3566 S = (Elf_Addr)S_tmp;
3568 stableVal = deRefStablePtr( stablePtr );
3570 S = (Elf_Addr)S_tmp;
3574 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3577 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p\n", symbol, (void*)S ));
3580 IF_DEBUG(linker,debugBelch( "Reloc: P = %p S = %p A = %p\n",
3581 (void*)P, (void*)S, (void*)A ));
3582 checkProddableBlock ( oc, pP );
3586 switch (ELF_R_TYPE(info)) {
3587 # ifdef i386_HOST_ARCH
3588 case R_386_32: *pP = value; break;
3589 case R_386_PC32: *pP = value - P; break;
3592 errorBelch("%s: unhandled ELF relocation(Rel) type %lu\n",
3593 oc->fileName, (lnat)ELF_R_TYPE(info));
3601 /* Do ELF relocations for which explicit addends are supplied.
3602 sparc-solaris relocations appear to be of this form. */
3604 do_Elf_Rela_relocations ( ObjectCode* oc, char* ehdrC,
3605 Elf_Shdr* shdr, int shnum,
3606 Elf_Sym* stab, char* strtab )
3609 char *symbol = NULL;
3611 Elf_Rela* rtab = (Elf_Rela*) (ehdrC + shdr[shnum].sh_offset);
3612 int nent = shdr[shnum].sh_size / sizeof(Elf_Rela);
3613 int target_shndx = shdr[shnum].sh_info;
3614 int symtab_shndx = shdr[shnum].sh_link;
3616 stab = (Elf_Sym*) (ehdrC + shdr[ symtab_shndx ].sh_offset);
3617 targ = (Elf_Addr) (ehdrC + shdr[ target_shndx ].sh_offset);
3618 IF_DEBUG(linker,debugBelch( "relocations for section %d using symtab %d\n",
3619 target_shndx, symtab_shndx ));
3621 for (j = 0; j < nent; j++) {
3622 #if defined(DEBUG) || defined(sparc_HOST_ARCH) || defined(ia64_HOST_ARCH) || defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
3623 /* This #ifdef only serves to avoid unused-var warnings. */
3624 Elf_Addr offset = rtab[j].r_offset;
3625 Elf_Addr P = targ + offset;
3627 Elf_Addr info = rtab[j].r_info;
3628 Elf_Addr A = rtab[j].r_addend;
3632 # if defined(sparc_HOST_ARCH)
3633 Elf_Word* pP = (Elf_Word*)P;
3635 # elif defined(powerpc_HOST_ARCH)
3639 IF_DEBUG(linker,debugBelch( "Rel entry %3d is raw(%6p %6p %6p) ",
3640 j, (void*)offset, (void*)info,
3643 IF_DEBUG(linker,debugBelch( " ZERO" ));
3646 Elf_Sym sym = stab[ELF_R_SYM(info)];
3647 /* First see if it is a local symbol. */
3648 if (ELF_ST_BIND(sym.st_info) == STB_LOCAL) {
3649 /* Yes, so we can get the address directly from the ELF symbol
3651 symbol = sym.st_name==0 ? "(noname)" : strtab+sym.st_name;
3653 (ehdrC + shdr[ sym.st_shndx ].sh_offset
3654 + stab[ELF_R_SYM(info)].st_value);
3655 #ifdef ELF_FUNCTION_DESC
3656 /* Make a function descriptor for this function */
3657 if (S && ELF_ST_TYPE(sym.st_info) == STT_FUNC) {
3658 S = allocateFunctionDesc(S + A);
3663 /* No, so look up the name in our global table. */
3664 symbol = strtab + sym.st_name;
3665 S_tmp = lookupSymbol( symbol );
3666 S = (Elf_Addr)S_tmp;
3668 #ifdef ELF_FUNCTION_DESC
3669 /* If a function, already a function descriptor - we would
3670 have to copy it to add an offset. */
3671 if (S && (ELF_ST_TYPE(sym.st_info) == STT_FUNC) && (A != 0))
3672 errorBelch("%s: function %s with addend %p", oc->fileName, symbol, (void *)A);
3676 errorBelch("%s: unknown symbol `%s'", oc->fileName, symbol);
3679 IF_DEBUG(linker,debugBelch( "`%s' resolves to %p", symbol, (void*)S ));
3682 IF_DEBUG(linker,debugBelch("Reloc: P = %p S = %p A = %p\n",
3683 (void*)P, (void*)S, (void*)A ));
3684 /* checkProddableBlock ( oc, (void*)P ); */
3688 switch (ELF_R_TYPE(info)) {
3689 # if defined(sparc_HOST_ARCH)
3690 case R_SPARC_WDISP30:
3691 w1 = *pP & 0xC0000000;
3692 w2 = (Elf_Word)((value - P) >> 2);
3693 ASSERT((w2 & 0xC0000000) == 0);
3698 w1 = *pP & 0xFFC00000;
3699 w2 = (Elf_Word)(value >> 10);
3700 ASSERT((w2 & 0xFFC00000) == 0);
3706 w2 = (Elf_Word)(value & 0x3FF);
3707 ASSERT((w2 & ~0x3FF) == 0);
3712 /* According to the Sun documentation:
3714 This relocation type resembles R_SPARC_32, except it refers to an
3715 unaligned word. That is, the word to be relocated must be treated
3716 as four separate bytes with arbitrary alignment, not as a word
3717 aligned according to the architecture requirements.
3720 w2 = (Elf_Word)value;
3722 // SPARC doesn't do misaligned writes of 32 bit words,
3723 // so we have to do this one byte-at-a-time.
3724 char *pPc = (char*)pP;
3725 pPc[0] = (char) ((Elf_Word)(w2 & 0xff000000) >> 24);
3726 pPc[1] = (char) ((Elf_Word)(w2 & 0x00ff0000) >> 16);
3727 pPc[2] = (char) ((Elf_Word)(w2 & 0x0000ff00) >> 8);
3728 pPc[3] = (char) ((Elf_Word)(w2 & 0x000000ff));
3732 w2 = (Elf_Word)value;
3735 # elif defined(powerpc_HOST_ARCH)
3736 case R_PPC_ADDR16_LO:
3737 *(Elf32_Half*) P = value;
3740 case R_PPC_ADDR16_HI:
3741 *(Elf32_Half*) P = value >> 16;
3744 case R_PPC_ADDR16_HA:
3745 *(Elf32_Half*) P = (value + 0x8000) >> 16;
3749 *(Elf32_Word *) P = value;
3753 *(Elf32_Word *) P = value - P;
3759 if( delta << 6 >> 6 != delta )
3761 value = (Elf_Addr) (&makeSymbolExtra( oc, ELF_R_SYM(info), value )
3765 if( value == 0 || delta << 6 >> 6 != delta )
3767 barf( "Unable to make SymbolExtra for #%d",
3773 *(Elf_Word *) P = (*(Elf_Word *) P & 0xfc000003)
3774 | (delta & 0x3fffffc);
3778 #if x86_64_HOST_ARCH
3780 *(Elf64_Xword *)P = value;
3785 #if defined(ALWAYS_PIC)
3786 barf("R_X86_64_PC32 relocation, but ALWAYS_PIC.");
3788 StgInt64 off = value - P;
3789 if (off >= 0x7fffffffL || off < -0x80000000L) {
3790 #if X86_64_ELF_NONPIC_HACK
3791 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3793 off = pltAddress + A - P;
3795 barf("R_X86_64_PC32 relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3796 symbol, off, oc->fileName );
3799 *(Elf64_Word *)P = (Elf64_Word)off;
3806 StgInt64 off = value - P;
3807 *(Elf64_Word *)P = (Elf64_Word)off;
3812 #if defined(ALWAYS_PIC)
3813 barf("R_X86_64_32 relocation, but ALWAYS_PIC.");
3815 if (value >= 0x7fffffffL) {
3816 #if X86_64_ELF_NONPIC_HACK
3817 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3819 value = pltAddress + A;
3821 barf("R_X86_64_32 relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3822 symbol, value, oc->fileName );
3825 *(Elf64_Word *)P = (Elf64_Word)value;
3830 #if defined(ALWAYS_PIC)
3831 barf("R_X86_64_32S relocation, but ALWAYS_PIC.");
3833 if ((StgInt64)value > 0x7fffffffL || (StgInt64)value < -0x80000000L) {
3834 #if X86_64_ELF_NONPIC_HACK
3835 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3837 value = pltAddress + A;
3839 barf("R_X86_64_32S relocation out of range: %s = %p\nRecompile %s with -fPIC.",
3840 symbol, value, oc->fileName );
3843 *(Elf64_Sword *)P = (Elf64_Sword)value;
3847 case R_X86_64_GOTPCREL:
3849 StgInt64 gotAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)->addr;
3850 StgInt64 off = gotAddress + A - P;
3851 *(Elf64_Word *)P = (Elf64_Word)off;
3855 case R_X86_64_PLT32:
3857 #if defined(ALWAYS_PIC)
3858 barf("R_X86_64_PLT32 relocation, but ALWAYS_PIC.");
3860 StgInt64 off = value - P;
3861 if (off >= 0x7fffffffL || off < -0x80000000L) {
3862 StgInt64 pltAddress = (StgInt64) &makeSymbolExtra(oc, ELF_R_SYM(info), S)
3864 off = pltAddress + A - P;
3866 *(Elf64_Word *)P = (Elf64_Word)off;
3873 errorBelch("%s: unhandled ELF relocation(RelA) type %lu\n",
3874 oc->fileName, (lnat)ELF_R_TYPE(info));
3883 ocResolve_ELF ( ObjectCode* oc )
3887 Elf_Sym* stab = NULL;
3888 char* ehdrC = (char*)(oc->image);
3889 Elf_Ehdr* ehdr = (Elf_Ehdr*) ehdrC;
3890 Elf_Shdr* shdr = (Elf_Shdr*) (ehdrC + ehdr->e_shoff);
3892 /* first find "the" symbol table */
3893 stab = (Elf_Sym*) findElfSection ( ehdrC, SHT_SYMTAB );
3895 /* also go find the string table */
3896 strtab = findElfSection ( ehdrC, SHT_STRTAB );
3898 if (stab == NULL || strtab == NULL) {
3899 errorBelch("%s: can't find string or symbol table", oc->fileName);
3903 /* Process the relocation sections. */
3904 for (shnum = 0; shnum < ehdr->e_shnum; shnum++) {
3905 if (shdr[shnum].sh_type == SHT_REL) {
3906 ok = do_Elf_Rel_relocations ( oc, ehdrC, shdr,
3907 shnum, stab, strtab );
3911 if (shdr[shnum].sh_type == SHT_RELA) {
3912 ok = do_Elf_Rela_relocations ( oc, ehdrC, shdr,
3913 shnum, stab, strtab );
3918 #if defined(powerpc_HOST_ARCH)
3919 ocFlushInstructionCache( oc );
3926 * PowerPC & X86_64 ELF specifics
3929 #if defined(powerpc_HOST_ARCH) || defined(x86_64_HOST_ARCH)
3931 static int ocAllocateSymbolExtras_ELF( ObjectCode *oc )
3937 ehdr = (Elf_Ehdr *) oc->image;
3938 shdr = (Elf_Shdr *) ( ((char *)oc->image) + ehdr->e_shoff );
3940 for( i = 0; i < ehdr->e_shnum; i++ )
3941 if( shdr[i].sh_type == SHT_SYMTAB )
3944 if( i == ehdr->e_shnum )
3946 errorBelch( "This ELF file contains no symtab" );
3950 if( shdr[i].sh_entsize != sizeof( Elf_Sym ) )
3952 errorBelch( "The entry size (%d) of the symtab isn't %d\n",
3953 (int) shdr[i].sh_entsize, (int) sizeof( Elf_Sym ) );
3958 return ocAllocateSymbolExtras( oc, shdr[i].sh_size / sizeof( Elf_Sym ), 0 );
3961 #endif /* powerpc */
3965 /* --------------------------------------------------------------------------
3967 * ------------------------------------------------------------------------*/
3969 #if defined(OBJFORMAT_MACHO)
3972 Support for MachO linking on Darwin/MacOS X
3973 by Wolfgang Thaller (wolfgang.thaller@gmx.net)
3975 I hereby formally apologize for the hackish nature of this code.
3976 Things that need to be done:
3977 *) implement ocVerifyImage_MachO
3978 *) add still more sanity checks.
3981 #if x86_64_HOST_ARCH || powerpc64_HOST_ARCH
3982 #define mach_header mach_header_64
3983 #define segment_command segment_command_64
3984 #define section section_64
3985 #define nlist nlist_64
3988 #ifdef powerpc_HOST_ARCH
3989 static int ocAllocateSymbolExtras_MachO(ObjectCode* oc)
3991 struct mach_header *header = (struct mach_header *) oc->image;
3992 struct load_command *lc = (struct load_command *) (header + 1);
3995 for( i = 0; i < header->ncmds; i++ )
3997 if( lc->cmd == LC_SYMTAB )
3999 // Find out the first and last undefined external
4000 // symbol, so we don't have to allocate too many
4002 struct symtab_command *symLC = (struct symtab_command *) lc;
4003 unsigned min = symLC->nsyms, max = 0;
4004 struct nlist *nlist =
4005 symLC ? (struct nlist*) ((char*) oc->image + symLC->symoff)
4007 for(i=0;i<symLC->nsyms;i++)
4009 if(nlist[i].n_type & N_STAB)
4011 else if(nlist[i].n_type & N_EXT)
4013 if((nlist[i].n_type & N_TYPE) == N_UNDF
4014 && (nlist[i].n_value == 0))
4024 return ocAllocateSymbolExtras(oc, max - min + 1, min);
4029 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
4031 return ocAllocateSymbolExtras(oc,0,0);
4034 #ifdef x86_64_HOST_ARCH
4035 static int ocAllocateSymbolExtras_MachO(ObjectCode* oc)
4037 struct mach_header *header = (struct mach_header *) oc->image;
4038 struct load_command *lc = (struct load_command *) (header + 1);
4041 for( i = 0; i < header->ncmds; i++ )
4043 if( lc->cmd == LC_SYMTAB )
4045 // Just allocate one entry for every symbol
4046 struct symtab_command *symLC = (struct symtab_command *) lc;
4048 return ocAllocateSymbolExtras(oc, symLC->nsyms, 0);
4051 lc = (struct load_command *) ( ((char *)lc) + lc->cmdsize );
4053 return ocAllocateSymbolExtras(oc,0,0);
4057 static int ocVerifyImage_MachO(ObjectCode* oc)
4059 char *image = (char*) oc->image;
4060 struct mach_header *header = (struct mach_header*) image;
4062 #if x86_64_HOST_ARCH || powerpc64_HOST_ARCH
4063 if(header->magic != MH_MAGIC_64) {
4064 errorBelch("%s: Bad magic. Expected: %08x, got: %08x.\n",
4065 oc->fileName, MH_MAGIC_64, header->magic);
4069 if(header->magic != MH_MAGIC) {
4070 errorBelch("%s: Bad magic. Expected: %08x, got: %08x.\n",
4071 oc->fileName, MH_MAGIC, header->magic);
4075 // FIXME: do some more verifying here
4079 static int resolveImports(
4082 struct symtab_command *symLC,
4083 struct section *sect, // ptr to lazy or non-lazy symbol pointer section
4084 unsigned long *indirectSyms,
4085 struct nlist *nlist)
4088 size_t itemSize = 4;
4091 int isJumpTable = 0;
4092 if(!strcmp(sect->sectname,"__jump_table"))
4096 ASSERT(sect->reserved2 == itemSize);
4100 for(i=0; i*itemSize < sect->size;i++)
4102 // according to otool, reserved1 contains the first index into the indirect symbol table
4103 struct nlist *symbol = &nlist[indirectSyms[sect->reserved1+i]];
4104 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4107 if((symbol->n_type & N_TYPE) == N_UNDF
4108 && (symbol->n_type & N_EXT) && (symbol->n_value != 0))
4109 addr = (void*) (symbol->n_value);
4111 addr = lookupSymbol(nm);
4114 errorBelch("\n%s: unknown symbol `%s'", oc->fileName, nm);
4122 checkProddableBlock(oc,image + sect->offset + i*itemSize);
4123 *(image + sect->offset + i*itemSize) = 0xe9; // jmp
4124 *(unsigned*)(image + sect->offset + i*itemSize + 1)
4125 = (char*)addr - (image + sect->offset + i*itemSize + 5);
4130 checkProddableBlock(oc,((void**)(image + sect->offset)) + i);
4131 ((void**)(image + sect->offset))[i] = addr;
4138 static unsigned long relocateAddress(
4141 struct section* sections,
4142 unsigned long address)
4145 for(i = 0; i < nSections; i++)
4147 if(sections[i].addr <= address
4148 && address < sections[i].addr + sections[i].size)
4150 return (unsigned long)oc->image
4151 + sections[i].offset + address - sections[i].addr;
4154 barf("Invalid Mach-O file:"
4155 "Address out of bounds while relocating object file");
4159 static int relocateSection(
4162 struct symtab_command *symLC, struct nlist *nlist,
4163 int nSections, struct section* sections, struct section *sect)
4165 struct relocation_info *relocs;
4168 if(!strcmp(sect->sectname,"__la_symbol_ptr"))
4170 else if(!strcmp(sect->sectname,"__nl_symbol_ptr"))
4172 else if(!strcmp(sect->sectname,"__la_sym_ptr2"))
4174 else if(!strcmp(sect->sectname,"__la_sym_ptr3"))
4178 relocs = (struct relocation_info*) (image + sect->reloff);
4182 #ifdef x86_64_HOST_ARCH
4183 struct relocation_info *reloc = &relocs[i];
4185 char *thingPtr = image + sect->offset + reloc->r_address;
4187 /* We shouldn't need to initialise this, but gcc on OS X 64 bit
4188 complains that it may be used uninitialized if we don't */
4191 int type = reloc->r_type;
4193 checkProddableBlock(oc,thingPtr);
4194 switch(reloc->r_length)
4197 thing = *(uint8_t*)thingPtr;
4198 baseValue = (uint64_t)thingPtr + 1;
4201 thing = *(uint16_t*)thingPtr;
4202 baseValue = (uint64_t)thingPtr + 2;
4205 thing = *(uint32_t*)thingPtr;
4206 baseValue = (uint64_t)thingPtr + 4;
4209 thing = *(uint64_t*)thingPtr;
4210 baseValue = (uint64_t)thingPtr + 8;
4213 barf("Unknown size.");
4216 if(type == X86_64_RELOC_GOT
4217 || type == X86_64_RELOC_GOT_LOAD)
4219 ASSERT(reloc->r_extern);
4220 value = (uint64_t) &makeSymbolExtra(oc, reloc->r_symbolnum, value)->addr;
4222 type = X86_64_RELOC_SIGNED;
4224 else if(reloc->r_extern)
4226 struct nlist *symbol = &nlist[reloc->r_symbolnum];
4227 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4228 if(symbol->n_value == 0)
4229 value = (uint64_t) lookupSymbol(nm);
4231 value = relocateAddress(oc, nSections, sections,
4236 value = sections[reloc->r_symbolnum-1].offset
4237 - sections[reloc->r_symbolnum-1].addr
4241 if(type == X86_64_RELOC_BRANCH)
4243 if((int32_t)(value - baseValue) != (int64_t)(value - baseValue))
4245 ASSERT(reloc->r_extern);
4246 value = (uint64_t) &makeSymbolExtra(oc, reloc->r_symbolnum, value)
4249 ASSERT((int32_t)(value - baseValue) == (int64_t)(value - baseValue));
4250 type = X86_64_RELOC_SIGNED;
4255 case X86_64_RELOC_UNSIGNED:
4256 ASSERT(!reloc->r_pcrel);
4259 case X86_64_RELOC_SIGNED:
4260 case X86_64_RELOC_SIGNED_1:
4261 case X86_64_RELOC_SIGNED_2:
4262 case X86_64_RELOC_SIGNED_4:
4263 ASSERT(reloc->r_pcrel);
4264 thing += value - baseValue;
4266 case X86_64_RELOC_SUBTRACTOR:
4267 ASSERT(!reloc->r_pcrel);
4271 barf("unkown relocation");
4274 switch(reloc->r_length)
4277 *(uint8_t*)thingPtr = thing;
4280 *(uint16_t*)thingPtr = thing;
4283 *(uint32_t*)thingPtr = thing;
4286 *(uint64_t*)thingPtr = thing;
4290 if(relocs[i].r_address & R_SCATTERED)
4292 struct scattered_relocation_info *scat =
4293 (struct scattered_relocation_info*) &relocs[i];
4297 if(scat->r_length == 2)
4299 unsigned long word = 0;
4300 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + scat->r_address);
4301 checkProddableBlock(oc,wordPtr);
4303 // Note on relocation types:
4304 // i386 uses the GENERIC_RELOC_* types,
4305 // while ppc uses special PPC_RELOC_* types.
4306 // *_RELOC_VANILLA and *_RELOC_PAIR have the same value
4307 // in both cases, all others are different.
4308 // Therefore, we use GENERIC_RELOC_VANILLA
4309 // and GENERIC_RELOC_PAIR instead of the PPC variants,
4310 // and use #ifdefs for the other types.
4312 // Step 1: Figure out what the relocated value should be
4313 if(scat->r_type == GENERIC_RELOC_VANILLA)
4315 word = *wordPtr + (unsigned long) relocateAddress(
4322 #ifdef powerpc_HOST_ARCH
4323 else if(scat->r_type == PPC_RELOC_SECTDIFF
4324 || scat->r_type == PPC_RELOC_LO16_SECTDIFF
4325 || scat->r_type == PPC_RELOC_HI16_SECTDIFF
4326 || scat->r_type == PPC_RELOC_HA16_SECTDIFF
4327 || scat->r_type == PPC_RELOC_LOCAL_SECTDIFF)
4329 else if(scat->r_type == GENERIC_RELOC_SECTDIFF
4330 || scat->r_type == GENERIC_RELOC_LOCAL_SECTDIFF)
4333 struct scattered_relocation_info *pair =
4334 (struct scattered_relocation_info*) &relocs[i+1];
4336 if(!pair->r_scattered || pair->r_type != GENERIC_RELOC_PAIR)
4337 barf("Invalid Mach-O file: "
4338 "RELOC_*_SECTDIFF not followed by RELOC_PAIR");
4340 word = (unsigned long)
4341 (relocateAddress(oc, nSections, sections, scat->r_value)
4342 - relocateAddress(oc, nSections, sections, pair->r_value));
4345 #ifdef powerpc_HOST_ARCH
4346 else if(scat->r_type == PPC_RELOC_HI16
4347 || scat->r_type == PPC_RELOC_LO16
4348 || scat->r_type == PPC_RELOC_HA16
4349 || scat->r_type == PPC_RELOC_LO14)
4350 { // these are generated by label+offset things
4351 struct relocation_info *pair = &relocs[i+1];
4352 if((pair->r_address & R_SCATTERED) || pair->r_type != PPC_RELOC_PAIR)
4353 barf("Invalid Mach-O file: "
4354 "PPC_RELOC_* not followed by PPC_RELOC_PAIR");
4356 if(scat->r_type == PPC_RELOC_LO16)
4358 word = ((unsigned short*) wordPtr)[1];
4359 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4361 else if(scat->r_type == PPC_RELOC_LO14)
4363 barf("Unsupported Relocation: PPC_RELOC_LO14");
4364 word = ((unsigned short*) wordPtr)[1] & 0xFFFC;
4365 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4367 else if(scat->r_type == PPC_RELOC_HI16)
4369 word = ((unsigned short*) wordPtr)[1] << 16;
4370 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
4372 else if(scat->r_type == PPC_RELOC_HA16)
4374 word = ((unsigned short*) wordPtr)[1] << 16;
4375 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
4379 word += (unsigned long) relocateAddress(oc, nSections, sections, scat->r_value)
4387 barf ("Don't know how to handle this Mach-O "
4388 "scattered relocation entry: "
4389 "object file %s; entry type %ld; "
4391 oc->fileName, scat->r_type, scat->r_address);
4395 #ifdef powerpc_HOST_ARCH
4396 if(scat->r_type == GENERIC_RELOC_VANILLA
4397 || scat->r_type == PPC_RELOC_SECTDIFF)
4399 if(scat->r_type == GENERIC_RELOC_VANILLA
4400 || scat->r_type == GENERIC_RELOC_SECTDIFF
4401 || scat->r_type == GENERIC_RELOC_LOCAL_SECTDIFF)
4406 #ifdef powerpc_HOST_ARCH
4407 else if(scat->r_type == PPC_RELOC_LO16_SECTDIFF || scat->r_type == PPC_RELOC_LO16)
4409 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
4411 else if(scat->r_type == PPC_RELOC_HI16_SECTDIFF || scat->r_type == PPC_RELOC_HI16)
4413 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
4415 else if(scat->r_type == PPC_RELOC_HA16_SECTDIFF || scat->r_type == PPC_RELOC_HA16)
4417 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
4418 + ((word & (1<<15)) ? 1 : 0);
4424 barf("Can't handle Mach-O scattered relocation entry "
4425 "with this r_length tag: "
4426 "object file %s; entry type %ld; "
4427 "r_length tag %ld; address %#lx\n",
4428 oc->fileName, scat->r_type, scat->r_length,
4433 else /* scat->r_pcrel */
4435 barf("Don't know how to handle *PC-relative* Mach-O "
4436 "scattered relocation entry: "
4437 "object file %s; entry type %ld; address %#lx\n",
4438 oc->fileName, scat->r_type, scat->r_address);
4443 else /* !(relocs[i].r_address & R_SCATTERED) */
4445 struct relocation_info *reloc = &relocs[i];
4446 if(reloc->r_pcrel && !reloc->r_extern)
4449 if(reloc->r_length == 2)
4451 unsigned long word = 0;
4452 #ifdef powerpc_HOST_ARCH
4453 unsigned long jumpIsland = 0;
4454 long offsetToJumpIsland = 0xBADBAD42; // initialise to bad value
4455 // to avoid warning and to catch
4459 unsigned long* wordPtr = (unsigned long*) (image + sect->offset + reloc->r_address);
4460 checkProddableBlock(oc,wordPtr);
4462 if(reloc->r_type == GENERIC_RELOC_VANILLA)
4466 #ifdef powerpc_HOST_ARCH
4467 else if(reloc->r_type == PPC_RELOC_LO16)
4469 word = ((unsigned short*) wordPtr)[1];
4470 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF) << 16;
4472 else if(reloc->r_type == PPC_RELOC_HI16)
4474 word = ((unsigned short*) wordPtr)[1] << 16;
4475 word |= ((unsigned long) relocs[i+1].r_address & 0xFFFF);
4477 else if(reloc->r_type == PPC_RELOC_HA16)
4479 word = ((unsigned short*) wordPtr)[1] << 16;
4480 word += ((short)relocs[i+1].r_address & (short)0xFFFF);
4482 else if(reloc->r_type == PPC_RELOC_BR24)
4485 word = (word & 0x03FFFFFC) | ((word & 0x02000000) ? 0xFC000000 : 0);
4490 barf("Can't handle this Mach-O relocation entry "
4492 "object file %s; entry type %ld; address %#lx\n",
4493 oc->fileName, reloc->r_type, reloc->r_address);
4497 if(!reloc->r_extern)
4500 sections[reloc->r_symbolnum-1].offset
4501 - sections[reloc->r_symbolnum-1].addr
4508 struct nlist *symbol = &nlist[reloc->r_symbolnum];
4509 char *nm = image + symLC->stroff + symbol->n_un.n_strx;
4510 void *symbolAddress = lookupSymbol(nm);
4513 errorBelch("\nunknown symbol `%s'", nm);
4519 #ifdef powerpc_HOST_ARCH
4520 // In the .o file, this should be a relative jump to NULL
4521 // and we'll change it to a relative jump to the symbol
4522 ASSERT(word + reloc->r_address == 0);
4523 jumpIsland = (unsigned long)
4524 &makeSymbolExtra(oc,
4526 (unsigned long) symbolAddress)
4530 offsetToJumpIsland = word + jumpIsland
4531 - (((long)image) + sect->offset - sect->addr);
4534 word += (unsigned long) symbolAddress
4535 - (((long)image) + sect->offset - sect->addr);
4539 word += (unsigned long) symbolAddress;
4543 if(reloc->r_type == GENERIC_RELOC_VANILLA)
4548 #ifdef powerpc_HOST_ARCH
4549 else if(reloc->r_type == PPC_RELOC_LO16)
4551 ((unsigned short*) wordPtr)[1] = word & 0xFFFF;
4554 else if(reloc->r_type == PPC_RELOC_HI16)
4556 ((unsigned short*) wordPtr)[1] = (word >> 16) & 0xFFFF;
4559 else if(reloc->r_type == PPC_RELOC_HA16)
4561 ((unsigned short*) wordPtr)[1] = ((word >> 16) & 0xFFFF)
4562 + ((word & (1<<15)) ? 1 : 0);
4565 else if(reloc->r_type == PPC_RELOC_BR24)
4567 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
4569 // The branch offset is too large.
4570 // Therefore, we try to use a jump island.
4573 barf("unconditional relative branch out of range: "
4574 "no jump island available");
4577 word = offsetToJumpIsland;
4578 if((long)word > (long)0x01FFFFFF || (long)word < (long)0xFFE00000)
4579 barf("unconditional relative branch out of range: "
4580 "jump island out of range");
4582 *wordPtr = (*wordPtr & 0xFC000003) | (word & 0x03FFFFFC);
4589 barf("Can't handle Mach-O relocation entry (not scattered) "
4590 "with this r_length tag: "
4591 "object file %s; entry type %ld; "
4592 "r_length tag %ld; address %#lx\n",
4593 oc->fileName, reloc->r_type, reloc->r_length,
4603 static int ocGetNames_MachO(ObjectCode* oc)
4605 char *image = (char*) oc->image;
4606 struct mach_header *header = (struct mach_header*) image;
4607 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
4608 unsigned i,curSymbol = 0;
4609 struct segment_command *segLC = NULL;
4610 struct section *sections;
4611 struct symtab_command *symLC = NULL;
4612 struct nlist *nlist;
4613 unsigned long commonSize = 0;
4614 char *commonStorage = NULL;
4615 unsigned long commonCounter;
4617 for(i=0;i<header->ncmds;i++)
4619 if(lc->cmd == LC_SEGMENT || lc->cmd == LC_SEGMENT_64)
4620 segLC = (struct segment_command*) lc;
4621 else if(lc->cmd == LC_SYMTAB)
4622 symLC = (struct symtab_command*) lc;
4623 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
4626 sections = (struct section*) (segLC+1);
4627 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
4631 barf("ocGetNames_MachO: no segment load command");
4633 for(i=0;i<segLC->nsects;i++)
4635 if(sections[i].size == 0)
4638 if((sections[i].flags & SECTION_TYPE) == S_ZEROFILL)
4640 char * zeroFillArea = stgCallocBytes(1,sections[i].size,
4641 "ocGetNames_MachO(common symbols)");
4642 sections[i].offset = zeroFillArea - image;
4645 if(!strcmp(sections[i].sectname,"__text"))
4646 addSection(oc, SECTIONKIND_CODE_OR_RODATA,
4647 (void*) (image + sections[i].offset),
4648 (void*) (image + sections[i].offset + sections[i].size));
4649 else if(!strcmp(sections[i].sectname,"__const"))
4650 addSection(oc, SECTIONKIND_RWDATA,
4651 (void*) (image + sections[i].offset),
4652 (void*) (image + sections[i].offset + sections[i].size));
4653 else if(!strcmp(sections[i].sectname,"__data"))
4654 addSection(oc, SECTIONKIND_RWDATA,
4655 (void*) (image + sections[i].offset),
4656 (void*) (image + sections[i].offset + sections[i].size));
4657 else if(!strcmp(sections[i].sectname,"__bss")
4658 || !strcmp(sections[i].sectname,"__common"))
4659 addSection(oc, SECTIONKIND_RWDATA,
4660 (void*) (image + sections[i].offset),
4661 (void*) (image + sections[i].offset + sections[i].size));
4663 addProddableBlock(oc, (void*) (image + sections[i].offset),
4667 // count external symbols defined here
4671 for(i=0;i<symLC->nsyms;i++)
4673 if(nlist[i].n_type & N_STAB)
4675 else if(nlist[i].n_type & N_EXT)
4677 if((nlist[i].n_type & N_TYPE) == N_UNDF
4678 && (nlist[i].n_value != 0))
4680 commonSize += nlist[i].n_value;
4683 else if((nlist[i].n_type & N_TYPE) == N_SECT)
4688 oc->symbols = stgMallocBytes(oc->n_symbols * sizeof(char*),
4689 "ocGetNames_MachO(oc->symbols)");
4693 for(i=0;i<symLC->nsyms;i++)
4695 if(nlist[i].n_type & N_STAB)
4697 else if((nlist[i].n_type & N_TYPE) == N_SECT)
4699 if(nlist[i].n_type & N_EXT)
4701 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
4702 if((nlist[i].n_desc & N_WEAK_DEF) && lookupSymbol(nm))
4703 ; // weak definition, and we already have a definition
4706 ghciInsertStrHashTable(oc->fileName, symhash, nm,
4708 + sections[nlist[i].n_sect-1].offset
4709 - sections[nlist[i].n_sect-1].addr
4710 + nlist[i].n_value);
4711 oc->symbols[curSymbol++] = nm;
4718 commonStorage = stgCallocBytes(1,commonSize,"ocGetNames_MachO(common symbols)");
4719 commonCounter = (unsigned long)commonStorage;
4722 for(i=0;i<symLC->nsyms;i++)
4724 if((nlist[i].n_type & N_TYPE) == N_UNDF
4725 && (nlist[i].n_type & N_EXT) && (nlist[i].n_value != 0))
4727 char *nm = image + symLC->stroff + nlist[i].n_un.n_strx;
4728 unsigned long sz = nlist[i].n_value;
4730 nlist[i].n_value = commonCounter;
4732 ghciInsertStrHashTable(oc->fileName, symhash, nm,
4733 (void*)commonCounter);
4734 oc->symbols[curSymbol++] = nm;
4736 commonCounter += sz;
4743 static int ocResolve_MachO(ObjectCode* oc)
4745 char *image = (char*) oc->image;
4746 struct mach_header *header = (struct mach_header*) image;
4747 struct load_command *lc = (struct load_command*) (image + sizeof(struct mach_header));
4749 struct segment_command *segLC = NULL;
4750 struct section *sections;
4751 struct symtab_command *symLC = NULL;
4752 struct dysymtab_command *dsymLC = NULL;
4753 struct nlist *nlist;
4755 for(i=0;i<header->ncmds;i++)
4757 if(lc->cmd == LC_SEGMENT || lc->cmd == LC_SEGMENT_64)
4758 segLC = (struct segment_command*) lc;
4759 else if(lc->cmd == LC_SYMTAB)
4760 symLC = (struct symtab_command*) lc;
4761 else if(lc->cmd == LC_DYSYMTAB)
4762 dsymLC = (struct dysymtab_command*) lc;
4763 lc = (struct load_command *) ( ((char*)lc) + lc->cmdsize );
4766 sections = (struct section*) (segLC+1);
4767 nlist = symLC ? (struct nlist*) (image + symLC->symoff)
4772 unsigned long *indirectSyms
4773 = (unsigned long*) (image + dsymLC->indirectsymoff);
4775 for(i=0;i<segLC->nsects;i++)
4777 if( !strcmp(sections[i].sectname,"__la_symbol_ptr")
4778 || !strcmp(sections[i].sectname,"__la_sym_ptr2")
4779 || !strcmp(sections[i].sectname,"__la_sym_ptr3"))
4781 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4784 else if(!strcmp(sections[i].sectname,"__nl_symbol_ptr")
4785 || !strcmp(sections[i].sectname,"__pointers"))
4787 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4790 else if(!strcmp(sections[i].sectname,"__jump_table"))
4792 if(!resolveImports(oc,image,symLC,§ions[i],indirectSyms,nlist))
4798 for(i=0;i<segLC->nsects;i++)
4800 if(!relocateSection(oc,image,symLC,nlist,segLC->nsects,sections,§ions[i]))
4804 #if defined (powerpc_HOST_ARCH)
4805 ocFlushInstructionCache( oc );
4811 #ifdef powerpc_HOST_ARCH
4813 * The Mach-O object format uses leading underscores. But not everywhere.
4814 * There is a small number of runtime support functions defined in
4815 * libcc_dynamic.a whose name does not have a leading underscore.
4816 * As a consequence, we can't get their address from C code.
4817 * We have to use inline assembler just to take the address of a function.
4821 extern void* symbolsWithoutUnderscore[];
4823 static void machoInitSymbolsWithoutUnderscore()
4825 void **p = symbolsWithoutUnderscore;
4826 __asm__ volatile(".globl _symbolsWithoutUnderscore\n.data\n_symbolsWithoutUnderscore:");
4828 #undef SymI_NeedsProto
4829 #define SymI_NeedsProto(x) \
4830 __asm__ volatile(".long " # x);
4832 RTS_MACHO_NOUNDERLINE_SYMBOLS
4834 __asm__ volatile(".text");
4836 #undef SymI_NeedsProto
4837 #define SymI_NeedsProto(x) \
4838 ghciInsertStrHashTable("(GHCi built-in symbols)", symhash, #x, *p++);
4840 RTS_MACHO_NOUNDERLINE_SYMBOLS
4842 #undef SymI_NeedsProto
4848 * Figure out by how much to shift the entire Mach-O file in memory
4849 * when loading so that its single segment ends up 16-byte-aligned
4851 static int machoGetMisalignment( FILE * f )
4853 struct mach_header header;
4856 fread(&header, sizeof(header), 1, f);
4859 #if x86_64_HOST_ARCH || powerpc64_HOST_ARCH
4860 if(header.magic != MH_MAGIC_64) {
4861 errorBelch("Bad magic. Expected: %08x, got: %08x.\n",
4862 MH_MAGIC_64, header->magic);
4866 if(header.magic != MH_MAGIC) {
4867 errorBelch("Bad magic. Expected: %08x, got: %08x.\n",
4868 MH_MAGIC, header->magic);
4873 misalignment = (header.sizeofcmds + sizeof(header))
4876 return misalignment ? (16 - misalignment) : 0;